Method and device for filtering a signal

Abstract

The present invention relates to a method and a device for filtering of a method* in which the following steps are carried out: a) furnishing a first signal value S1†, a second signal value and a third signal value, b) determining an interval dictated by the at least one second signal value and the at least one third signal value, with a first interval boundary depending on the at least one second signal value and a second interval boundary dictated by the at least one third signal value, c) determining a mean value from the at least one selected second signal value and the at least one selected third signal value, d) determining whether the first signal value lies within the interval, e) if the first signal value lies outside the interval: determining a first mathematical distance between the first signal value and the first interval boundary and a second mathematical distance between the first signal value and the second interval boundary and a third mathematical distance between‡ between the first signal value and the mean value to furnish a first, a second, and a third distance value, selecting a value from a value range between the smaller of the first and second distance values and the third distance value, f) if the first signal value lies within the interval: selection of a value from a value range between zero and the half-interval width, g) choosing the sign of the selected signal value depending on the position of the first signal value relative to the mean value to furnish the filtered signal value. * Sic. Presumably, “of a method” should be deleted. Translator. † Sic. “S1” should be deleted as there are no other reference characters in the abstract. Translator. ‡ Sic. The first “between” should be deleted. Translator.

Description

The present invention concerns a method and a device for filtering, especially for high-pass filtering, of a signal.

As represented in FIG. 15a, high-pass filters 200 are used in particular also for low-pass filtering of signals in(k), the high-pass filter 200 being used to filter out high-frequency signal components from the input signal in(k), in order to subtract these high-frequency signal components—possibly after amplification or attenuation by means of an amplifier 218—from the input signal in(k), so as to obtain a low-pass filtered signal as the output signal out(k). A delay element 215, shown in FIG. 15a, serves to equalize the signal delays introduced by the high-pass filter 200.

Traditional high-pass filters are linear filters, such as the FIR filter shown in FIG. 200, whose diagram is represented, and which possesses a transfer function h(z)=0.25·[−z⁻²+2z⁻¹−1].

With respect to the input signal in(k), such a filter provides an output signal value of three consecutive input signal values, the output signal value corresponding to a normalized difference between the middle of the three signal values and the mean value of the other two signal values.

The mode of operation of such a low-pass filter arrangement for an input signal in(k), represented as an example in FIG. 15b, is presented in FIG. 15c. This shows that the low-pass filter filters out high-frequency signal components, but also pulselike signal components are decreased and widened at their base, and moreover edges of the input signal are partly smoothed out, having the result that the high-pass filter also filters out the pulses and signal steps, which are then subtracted from the input signal.

The range of application of such filter arrangements is very large, and the discrete signals that are filtered are formed, for example, by sampling a signal that is continuous over time. However, such signals can also be graphical information values, which are taken from an image line by line or column by column. In the processing of television signals, there are known methods in which frames with graphical information in every line are generated from the transmitted fields, in which only every other line contains graphical information. In the event of an incorrect interpolation of the intermediate-image lines, it may happen that a signal variation is produced in the vertical direction of the picture, where maximum and minimum signal values alternate from line to line (for example, luminance values or chrominance values). Such high-frequency signal processes need to be filtered, so as for example to generate a gray value from a black/white variation from one line to another, in every line. However, individual signal steps, which are represented as brightness edges or color edges in the picture, should remain unaffected by such a filter process. The same holds for signal pulses, i.e., narrow bright or dark regions in the picture, which likewise should not be wiped out by the filter process.

Thus, the basic goal of the present invention is to provide a method and a device for the high-pass filtering of a signal, which is able to filter out high-frequency signal components but does not filter out signal pulses or signal steps, or does so only to a small degree.

This goal is accomplished by a method according to claim 1 and by a high-pass filter arrangement according to claim 14. Advantageous embodiments of the invention are the subjects of the dependent claims.

The method according to the invention for high-pass filtering of a signal having a succession of signal values, in order to provide a high-pass filtered signal having a succession of signal values, comprises the following steps in order to generate a signal value of the filtered signal each time:

• • a) selection of a first signal value of the signal to be filtered and at least one second signal value preceding the first signal value in the succession and at least one third signal value following the first signal value in the succession,
  • b) determination of an interval dictated by the at least one second signal value and the at least one third signal value, with a first interval boundary dependent on the at least one second signal value and a second interval boundary dictated by the at least one third interval value,
  • c) determination of a mean value from the at least one selected second signal value and the at least one selected third signal value,
  • d) determination as to whether the first signal value lies within the interval,
  • e) if the first signal value lies outside the interval: determination of a first mathematical distance between the first signal value and the first interval boundary and a second mathematical distance between the first signal value and the second interval boundary and a third mathematical distance between the first signal value and the mean value to furnish a first, second, and third distance value, selection of a value from a value range between the smaller of the first and second distance values and the third distance value,
  • f) if the first signal value lies within the interval: selection of a value from a value range between zero and half of the interval width,
  • g) setting the sign of the selected signal value depending on the position of the first signal value relative to the mean value, to provide the filtered signal value.

The method according to the invention for high-pass filtering is a nonlinear method, in which one first determines whether a first signal value lying in the signal sequence between the at least one second signal value and the at least one third signal value lies inside or outside an interval spanned by the second and third signal values. If this first signal value lies within this interval, then a filter signal value of zero is preferably generated. If the first signal values lies outside the interval boundaries formed by the at least one second signal value and the at least one third signal value, there is a pulselike deviation of the first signal value relative to these interval boundaries, so that a filtered signal value not equal to zero is generated, reflecting this pulselike deviation of this first signal value relative to the interval boundaries. The value of the filtered signal value is then preferably chosen so that it corresponds to the distance between the first signal value and the closest interval boundary.

The sign of the filtered signal value depends on the position of the first signal value relative to the mean of the interval boundaries, and for example the sign of the filtered signal value is set at minus when the first signal value is smaller than this mean, and this sign is preferably set at plus when this first signal value is larger than this mean.

In order to suppress noise during the high-pass filtering, it is preferable to set at zero the distance value resulting from the determination of the first, second, and third distance if the distance is smaller than a given limit value. This is based on the knowledge that slight deviations between signals, for example, between a first signal value and the closest interval boundaries formed by one of the neighboring signal values, usually result from noise factors, which can be suppressed by the method explained.

In an exemplary embodiment of the invention, the interval boundaries are formed by a second signal value preceding the first signal value in the sequence and a third signal value following the first signal value in the sequence. Another embodiment specifies how to select several second signal values preceding the first signal value and several third signal values following the first signal value, form an interval between each second signal value and each third signal value, and for example select as interval boundaries the second signal value and the third signal value that span the largest interval or the smallest interval. Basically, it is also possible to select any one of the second and third signal values.

Preferably, the filtered signal value is confined to an upper or lower limit value.

The method according to the invention is especially suitable for low-pass filtering in that it discloses how to produce a high-pass filtered signal according to the method according to the invention and how to generate the low-pass filtered signal from the high-pass filtered signal and the input signal.

In the simplest case, the high-pass filtered signal is subtracted from the delayed input signal in order to provide a low-pass filtered signal.

In another embodiment of a low-pass filter method applying the high-pass filter method according to the invention, the following method steps are provided:

• • limiting the high-pass filtered signal to generate a bounded high-pass filtered signal,
  • subtracting the bounded high-pass filtered signal from the input signal to generate a first low-pass filtered signal,
  • subtracting the bounded high-pass filtered signal from the high-pass filtered signal to generate a difference signal,
  • limiting a signal dependent on the difference signal,
  • adding the first low-pass filtered signal and the bounded difference signal to form the low-pass filtered signal.

In a modification of the last mentioned low-pass filter method, the difference signal is filtered prior to the limiting.

Furthermore, the possibility exists of weighting the difference signal and adding the weighted difference signal to the first low-pass filtered signal and to the bounded difference signal in order to generate a low-pass filtered output signal.

A high-pass filter arrangement according to the invention for filtering an input signal having a succession of signal values comprises an input for supplying the input signal and an output for providing a filter output signal, as well as a first and a second high-pass filter, each of which is connected behind the input. Furthermore, the high-pass filter arrangement comprises a coring unit with a signal input, which receives a signal depending on a filter output signal of the first high-pass filter, with a control input, which receives a signal depending on a signal at the output of the second high-pass filter, and with an output which is coupled to the output of the high-pass filter arrangement.

This high-pass filter arrangement according to the invention adaptively filters the input signal, in that the parameter of the coring unit is set each time depending on the filter output signal of the second high-pass filter, and the input signal is supplied to this second high-pass filter.

Preferably, between the output of the second high-pass filter and the control input of the coring unit an arrangement is connected to form the absolute value of the filter output signal of the second high-pass filter. Optionally, moreover, an amplifier arrangement is connected between the second high-pass filter and the control input of the coring unit, which preferably has a gain between 0 and 1, and which thus offers an opportunity to adjust the filter behavior.

The first high-pass filter preferably has a transfer function for which: H1(z)=k1·[−z⁻²+2z⁻¹−1], where k1 is a constant. Such a filter processes three consecutive signal values of the input signal and makes available at the output a filtered value that is proportional to the difference between the middle of the three signal values and the mean of the other two signal values.

Preferably, for the transfer function of the second high-pass filter: H2(z)=k2·[z⁻²−1], where k2 is a constant. A high-pass filter with such a transfer function processes three consecutive signal values and provides a filter value that is proportional to a difference between the first and third of the three consecutive input signal values. Referring back to the above-explained method according to the invention, this filter signal value corresponds to the interval between the second and third signal values.

The coring unit, as is known, is used to set at zero the input signal values that lie within an interval dictated by a control signal, while values that are smaller or larger than this interval are decreased or increased by the control signal value, regardless of whether the input signal value is larger than the upper interval boundary dictated by the control signal or smaller than the lower interval boundary dictated by the control signal value.

When two high-pass filters with the above-explained transfer functions are used, the control signal which represents the coring parameter of the coring unit is thus dependent on the interval width between the second and third signal values, and this control signal is applied to a signal corresponding to the difference between a first signal value and the mean between the second and third signal values. Preferably, a device is connected between the output of the coring unit and the output of the filter device so that the output signal of the coring unit can be adjustably amplified or attenuated from the outside.

Preferably, a limiting arrangement is connected between the output of the coring unit and the output of the filter arrangement.

The high-pass filter arrangement according to the invention is suitable for use in a low-pass filter arrangement, in which an input signal is combined by means of a switching circuit with the output signal of the high-pass filter arrangement in order to form a low-pass filtered signal.

The present invention will now be explained in greater detail hereinbelow in exemplary embodiments with reference to the drawings:

FIG. 1 shows three consecutive signal values of a signal to be filtered, to explain the method according to the invention,

FIG. 2 shows three additional consecutive signal values of a signal to be filtered, to explain the method according to the invention,

FIG. 3 shows an alternating signal to be filtered and the resulting filtered signal according to the invention,

FIG. 4 shows a step signal and the resulting filtered signal (FIG. 4a) and a pulse signal and the resulting filtered signal (FIG. 4b),

FIG. 5 shows an exemplary embodiment of a high-pass filter according to the invention,

FIG. 6 shows an exemplary embodiment of a low-pass filter arrangement with a high-pass filter according to the invention,

FIG. 7 shows the low-pass filter according to FIG. 6 with a more detailed representation of the components of the high-pass filter,

FIG. 8 is an exemplary diagram of the input signal (FIG. 8a) and the output signal (FIG. 8b) of the filter arrangement according to FIG. 7,

FIG. 9 is an exemplary time diagram of the filter input signal of the filter according to FIG. 7, the filter output signal in a high-pass filtering of only every other (even) input value (FIG. 9b) and the filter output signal in high-pass filtering of only every other (odd) input value (FIG. 9c),

FIG. 10 shows another exemplary embodiment of a low-pass filter with a high-pass filter arrangement according to the invention, wherein the high-pass filtered signal is limited at the output of the high-pass filter,

FIG. 11 shows another exemplary embodiment of a low-pass filter arrangement with a high-pass filter according to the invention,

FIG. 12 shows another exemplary embodiment of a low-pass filter arrangement with a high-pass filter according to the invention,

FIG. 13 shows another exemplary embodiment of a low-pass filter arrangement with a high-pass filter according to the invention,

FIG. 14 shows a filter system with a consecutive switching of a low-pass filter arrangement and a high-pass filter arrangement, and exemplary time diagrams of several signals occurring in the system,

FIG. 15 shows a modification of the filter device according to FIG. 7 for application of the method according to the invention to any signal values, not necessarily consecutive in time or space,

FIG. 16 is a schematic representation of a field sequence to explain the application of the method to a pixel interpolation,

FIG. 17 is a schematic representation of an image sequence and an intermediate image to be interpolated by use of motion vectors to explain the application of the method to a pixel interpolation assisted by motion vectors,

FIG. 18 shows a filter system according to the prior art (FIG. 15a), as well as an exemplary filter input signal and a corresponding filter output signal (FIGS. 15b, c),

FIG. 19 shows a filter arrangement with series connection of a low-pass filter and a high-pass filter according to the prior art.

In the figures, unless otherwise indicated, the same reference numbers designate the same parts and signals with identical meaning.

The method according to the invention for filtering a signal s(k) having a sequence of signal values in order to generate a filtered signal hp(k) likewise having a sequence of signal values is explained hereinbelow for the generation of a signal value hp(i) of the filtered signal hp(k) with reference to FIGS. 1 and 2. In the following, k each time designates the control variables of the discrete signals s(k), hp(k), each comprising a sequence of signal values, and i hereinbelow designates any given element of these control variables.

The signal s(k) to be filtered can be any signal having a sequence of signal values, in particular a time-discrete signal, and thus it can represent sample pulses of a time-continuous signal. The sequence of signal values s(k) also represents a sequence of graphical information values, which are read out column by column or line by line from a picture.

The method according to the invention, in order to furnish a signal value hp(i) of the filtered signal calls for selecting a signal value of the signal s(k) to be filtered, which is represented by the signal value s(i) in FIGS. 1 and 2. Furthermore, at least one second signal value preceding the selected first signal value s(i), being represented by the signal value s(i−1), and at least one signal value following the selected signal value s(i), represented in FIGS. 1 and 2 by the signal value s(i+1), are selected. The second and third signal values s(i−1), s(i+1) span an interval I=[s(i−1); s(i+1)], whose width is designated as t in FIGS. 1 and 2.

Besides this interval, the mean value of the first and second signal values s(i−1), s(i+1) is determined, being designated as m in FIGS. 1 and 2.

A determination is made as to whether the first signal value s(i) lies within the interval given by the second and third signal values s(i−1), s(i+1). If so (see FIG. 1), a first distance d1 is determined between the first signal value s(i) and the second signal value s(i−1), a second distance d2 between the first signal value s(i) and the third signal value s(i+1), and a third distance between the first signal value s(i) and the middle of the interval or the mean value m. The filtered signal value hp(i) is then selected from an interval which is spanned by the smaller of the first and second distance d1, d2 and the third distance d3, the sign of the filtered signal value depending on the position of the first signal value s(i) relative to the mean value m, such that the sign of the filtered signal value hp(i) is minus when the first signal value s(i) is smaller than the mean value m, as is represented in FIG. 1, and the sign of the filtered signal value hp(i) is plus when the first signal value s(i) is greater than the mean value m.

The selecting of the smaller of the first and second distance d1, d2 results in the magnitude of the filtered value hp(i) lying in an interval which is formed by the distance to the closest interval boundary and the distance to the middle of the interval. In the example of FIG. 1, the second signal value s(i−1) forms the closest interval boundary to the signal value s(i).

If the selected signal value s(i) should lie within the interval of width t spanned by the second signal value s(i−1) and the third signal value s(i+1), a signal value is chosen for the filtered signal value hp(i) extending within an interval from zero to half the interval width t/2, the sign of the filtered signal value hp(i) being minus if the first signal value is smaller than the mean value m, and plus if the first signal value s(i) is greater than the mean value m.

It follows from the above that, in the case when the second and third signal values s(i−1), s(i+1) are the same size, but the first signal value s(i) is greater or smaller than the other two signal values, a value is output as the filtered signal value hp(i) whose magnitude corresponds to the difference between the first signal value s(i) and one of the other two signal values s(i−1), s(i+1), and whose sign is chosen depending on whether the first signal value s(i) is greater or smaller than the other two signal values.

In the event that the first, second, and third signal value are the same size, the filtered signal value hp(i) is equal to zero.

In order to suppress noise influences, it is preferable to set at zero the first through third distances determined during the method and the interval width, provided the value determined is less than a given limit value, since one must assume that only slight deviations of the particular signal values will result from unwanted noise influences.

The method according to the invention will preferably output, as the filtered signal value hp(i), a value whose magnitude corresponds to the distance between the first signal value s(i) and the closest interval boundary and whose sign is chosen depending on the position of the first signal value s(i) relative to the mean value m, if the first signal value s(i) lies outside the interval spanned by the second and third signal values. Preferably, moreover, the filtered signal value hp(i) will be set at zero if the first signal value s(i) lies within the interval spanned by the second and third signal values s(i−1), s(i+1). In this special configuration of the method, the filtered signal value hp(i) is equal to zero if the first signal value s(i) corresponds to the second or third signal value s(i−1), s(i+1).

For purposes of explanation, it has been assumed thus far that the second and third signal values spanning the interval come immediately before and immediately after the particular signal value s(i) in the signal sequence, and these second and third signal values form the boundaries of the interval that is considered when generating the filtered signal value hp(i).

However, the possibility also exists of selecting several second signal values preceding the particular signal value s(i) and several third signal values following the particular signal value s(i) and forming an interval between each of the second signal values and each of the third signal values and selecting the first signal value and the third signal value that span the largest interval, this second and this third signal value then forming the boundaries of the interval considered thereafter.

Mathematically, the signal value hp(k) generated by the filtering in the case when the first signal value s(i) lies outside the interval formed by the second and third signal values s(i−1), s(i+1), can be formed by forming the difference between the first signal value s(i) and the mean value m, and subtracting the magnitude of half the interval width from the resulting difference if the signal value s(i) is larger than the upper interval boundary, and adding the magnitude of half the interval width to the resulting difference if the signal value s(i) is smaller than the lower interval boundary. If s(i) lies above the upper interval boundary, the difference determined between the signal value s(i) and the half-interval boundary is positive, and the value obtained by subtracting the half-interval width from this difference value is preferably bounded at the bottom as far as zero. If s(i) lies below the lower interval boundary, the difference determined between the signal value s(i) and the half-interval boundary is negative, and the value obtained by adding this difference value to the half-interval width is preferably bounded at the top by zero. Also, in this special procedure for determining a filtered value hp(k), both the difference and the half-interval width are preferably set at zero when the value determined for this difference or the half-interval width is smaller than a given limit value.

FIG. 3 illustrates the operation of the method according to the invention when filtering an alternating signal s(k), which alternately takes on signal values of −a and +a, for the special embodiment of the method where the particular filter signal value corresponds to the distance to the closest interval boundary, if the first signal value lies outside the interval formed by the second and third signal values, and where the filter signal value is set at zero if the first signal value lies within the interval. The resulting high-pass signal hp(k) is likewise an alternating signal that alternately takes on negative signal values with an amplitude of −2a and positive signal values with an amplitude of +2a, thus, the filtered signal corresponds to the input signal with twice the amplitude.

The special advantage of the method according to the invention becomes clear from FIG. 4a, in which a step signal s(k) is represented, which at first has a sequence of constant consecutive values (with zero amplitude in the example) and after an abrupt change it has a constant sequence of signal values with amplitude not equal to zero. The resulting high-pass signal hp(k), which has been formed by determining for each signal value the distance to the next interval boundary of an interval formed by the preceding and the following signal value, “ignores” this step and represents a constant signal with zero amplitude.

FIG. 4 b illustrates the high-pass filter method according to the invention by means of a signal s(k) having a pulse of amplitude a, wherein the filtered signal hp(k) likewise contains a pulse of amplitude a.

FIG. 5 shows an exemplary embodiment of a high-pass filter according to the invention for implementing the method according to the invention. This high-pass filter arrangement comprises a first high-pass filter 10 with a transfer function H1(z) and a second high-pass filter 20 with a transfer function H2(z), each of them receiving the signal s(k) for filtering at the input of the high-pass filter arrangement. A coring unit 12, which has a signal input to supply a signal for processing and a control input, is connected behind first and second filters 10, 20. The control input of this coring unit 12 receives an output signal p(k) from the second high-pass filter 20 and the signal input of the coring unit 12 receives an output signal hp1(k) of the first filter 10. The high-pass filtered signal hp(k) is available at the output of the coring unit. Optionally, the output signal of the second high-pass filter can be made to form the absolute value before being taken to the coring unit.

The mode of operation of a coring unit is sufficiently familiar and will be briefly summarized here. An output signal value hp(i) furnished by the coring unit 12 is zero if the input signal value hp1(i) lies within an interval given by −p(i) and +p(i). If the input signal value hp1(i) is greater than p(1), the output signal value hp(i) will be the difference between the input signal value and the control signal value p(i). If the input signal value hp1(i) is less than the negative control signal value −p(i), the output signal value hp(i) will be the sum of the input signal value hp1(i) and the control signal value p(i).

The transfer function H1(z) of the first high-pass filter is preferably: H1(z)=0.5·[−z⁻²+2·z⁻¹−1].

In terms of the discrete input sequence s(k), this means that a signal value hp1(i) at the output of the first filter 10 is proportional to a difference between a signal value s(i) and the mean value of the preceding value s(i−1) and following value s(i+1). For the transfer function H2(z) of the second filter 20, we have preferably: H2(z)=0.5·[1−z⁻²]  (1).

In terms of the discrete signal s(k), this means that the signal used as the parameter signal for the coring unit 12 represents each time, for a signal value s(i) of the discrete signal s(k), the half-interval width of an interval spanned by the signal value s(i−1) preceding this value s(i) and a signal value s(i+1) following this value. The output signal of the coring unit 12 is preferably subjected to amplification or damping before being taken to the output of the filter.

In this filter arrangement, therefore, a filter output value of zero is output if the difference between a signal value and the mean value of the neighboring signal values is less than the half-interval width, i.e., if this value lies within the interval spanned by the neighboring signal values. If the difference between the signal value and the mean value of the neighboring signal values is larger than the half-interval width represented by p(k), a filter signal value will be output that corresponds to the difference between the signal value and the mean value plus or minus the half-interval width, and which thus represents the distance between the signal value and the closest interval boundary of the interval spanned by the neighboring signal values.

As depicted by broken line in FIG. 5, an amplifier is optionally connected to the second high-pass filter 20, multiplying the output signal of the second filter with a gain factor v, and this gain factor v preferably lies in the region between zero and one.

By means of this gain factor v, the amplitude of the high-pass filtered signal hp(k) can be varied, where a gain factor of zero corresponds to the case in which the coring unit lets the signal present at the output of the first filter 10 pass through to the output unchanged, wherein the individual signal values in view of the above-mentioned transfer function of the first filter then correspond each time to the difference between a signal value and the mean value of the neighboring signal values.

The high-pass filter according to the invention in FIG. 5 is especially suitable for use in a low-pass filter, represented in FIG. 6. This low-pass filter contains, besides the high-pass filter according to the invention, only a subtractor 14, whose noninverting input is the input signal s(k) to be filtered and whose inverting input receive the high-pass filtered signal hp(k). Furthermore, a delay element is connected ahead of the noninverting input of the subtractor, in order to delay the signal s(k) to be filtered in accordance with the delays in the high-pass filter.

The high-pass filter signal hp(k) is weighted by means of a weighting arrangement 17 with a multiplication factor 0.5 prior to the subtraction of the delayed input signal s(k) in the low-pass filter arrangement in FIG. 6.

FIG. 7 shows the low-pass filter arrangement depicted in FIG. 6 with the first and second high-pass filters 10, 20 shown in detail. The first high-pass filter 10 has a transfer function H1(z), for which: H1(z)=0.5·[−z⁻²−2z⁻¹−1]  (2) and is constructed as an FIR filter with a sufficiently familiar nonrecursive structure made of weighting elements 101, 102, 103, delay elements 107, 108, adders 104, 105, and another amplification element 106 connected ahead of the filter output.

The second low-pass filter 20 has the transfer function H2(z) referred to above, and this filter as well is constructed as an FIR filter from a nonrecursive structure, which comprises a weighting element 201, delay elements 207, 208, an adder 204, and a weighting element 206 connected ahead of the output of the high-pass filter 20. Then, for a transfer function P(z), which defines the mapping of the input signal p(k) onto the signal p(k) supplied to the coring unit in the frequency domain, we have: P(z)=v·|0.5·(z⁻²−1)|  (3), where v is the gain factor of the amplifier 24, and this gain factor is preferably 1.

FIG. 7, furthermore, shows the mode of operation of the coring unit 12 by means of the transfer characteristic of this coring unit, making use of the signal p(k) present at the control input of control unit 12, which results from the filter output signal of the second high-pass filter 20 by forming the magnitude and weighting, and also by means of the filter output signal hp1(k) of the first high-pass filter 10 and by means of the high-pass filtered signal hp(k) present at the output of the coring unit. The characteristic makes it clear that the input signal hp1 is set at zero for value domains between −p(k) and +p(k), that it results from the difference of hp1(k) and p(k) for value domains larger than p(k), and that it results from the sum of hp1(k) and p(k) for values less than −p(k).

FIG. 8 a shows, as an example, one possible input signal s(k) fed to the low-pass filter arrangement in FIG. 7. FIG. 8b shows the low-pass filtered signal tp(k) resulting from this input signal according to FIG. 8 at the output of the filter arrangement according to FIG. 7. For better comprehension, points in the course of the input signal s(k) at which conspicuous changes in the signal occur from control variables k(0) to k(6) are highlighted in FIG. 8. Corresponding control variables k(0) through k(6) are represented in FIG. 8b, ignoring delays in passing through the low-pass filter arrangement, in order to emphasize the effects of the low-pass filtering at the places with significant signal changes.

The input signal s(k) in the depicted feature is a step signal with a step from a lower level to an upper level at k3, there being superimposed on this step signal a positive pulse at k0 and a negative pulse at k4, as well as a high-frequency signal component between k1 and k3 and k5 and k6.

As can be seen from the pattern of the low-pass filtered signal tp(k), the low-pass filter leaves the step, as well as the lower level and the upper level of the step function, unchanged, since in the manner explained the high-pass filtered signal hp(k) has a signal value of zero for constant values of the input signal and for a step in the input signal from a lower constant value to an upper constant value. The pulses at k0 and k4 remain essentially preserved, but are cut in half in their amplitude as compared to the ground level, which represents the lower level of the step signal for the pulse at k0, and the upper level of the step signal for the pulse at k4, since in the already explained manner the high-pass filter arrangement responds to a pulse likewise with a pulse, and this pulse of the high-pass filtered signal's high-pass signal hp(k) in FIG. 7 is subtracted from or added to the pulse of the input signal s(k) with a factor 0.5 with half its amplitude.

There is an optional possibility for varying the amplitude of the pulses remaining after the low-pass filtering tp(k), by weighting the high-pass filtered signal present at the output of the coring unit 12 by means of the weighting arrangement 18 depicted in FIG. 7, to provide the output signal hp(k) of the high-pass filter, and this weighting factor v2 preferably lies in an interval between 0 and 1. In the extreme case of a value of zero, the input signal s(k) is output unchanged, only delayed, as the low-pass filtered signal tp(k). In the other extreme case of a weighting factor v2 of 1, there is subtracted from the input signal s(k) a high-pass filtered signal, whose amplitude for pulses corresponds precisely to the amplitude of the pulses of the input signal s(k), so that pulses in the low-pass filtered signal tp(k) are entirely extinguished. With a gain factor v2=0.5, the amplitude of the pulses is cut in half. Summarizing, the method according to the invention for high-pass filtering leaves steps of the input signal s(k) unchanged, and furthermore by weighting the high-pass filtered signal hp(k) before subtraction from the input signal s(k) it is possible to adjust how much the pulses of the input signal s(k) should be damped in the low-pass filtered signal tp(k) present at the output of the low-pass filter arrangement tp(k).

As is further shown by FIG. 8b, the high-frequency signal components are largely extinguished, and pulses are only generated at the beginning and end of the particular high-frequency signal component.

FIG. 9 shows yet again the signal diagram in FIG. 8a, while in FIG. 9 this signal diagram is contrasted with signal diagrams in FIGS. 9b and 9c, which were obtained by subtracting each time only every other high-pass filtered value present at the output of the coring unit 12 from the input signal s(k) in the filter arrangement in FIG. 7. For this purpose, one can provide, for example, a sampling unit 1 as indicated by broken line in FIG. 7, which is connected to the coring unit and configured such that it sets at zero every other output value hp(k), i.e., either the values with even-numbered control variable k or the values with odd-numbered control variable k. In this way, only every other value of the input signal s(k) is low-pass filtered, while the other values arrive unchanged, merely delayed in time, at the output.

Depending on the application, however, the sampling pattern can be varied at will, for example by setting at zero only every n-th output value hp(k) or letting through only the n-th value, where n is an integer value.

FIG. 9 b shows the diagram obtained when only the values with even-numbered index k are low-pass filtered, where the pulses at k0 and k4 are not even contained in this sampled signal, subjected to the low-pass filtering, and are therefore not considered in the high-pass filtering, and thus are contained unchanged in the low-pass filtered signal tp(k). It is further clear that this type of filtering creates an envelope for the high-frequency signal components between k1, k2 and k5, k6.

FIG. 9 c shows the filter result for the case in which all input signal values s(k) with odd-numbered indices are low-pass filtered, these filtered signal values containing the pulses at k0 and k4, and these pulses are fully extinguished in the low-pass filtered signal tp(k) present at the output, provided the gain factors are v=1 and v2=1, since the high-pass filtered signal then contains the pulses at their full amplitude and these pulses are subtracted from the input signal.

FIG. 10 shows a low-pass filter arrangement which has been modified from FIGS. 6 and 7, containing a high-pass filter according to the invention, FIG. 5. In this low-pass filter arrangement, the high-pass filtered signal hp(k) present at the output of the high-pass filter arrangement is limited by a limiting unit 16 before the high-pass filtered signal hp(k) is subtracted from the delayed input signal s(k). The high-pass filtered signal in FIG. 10, and in the FIGS. 11 through 13 which are explained below, corresponds to the output signal of the coring unit 12, if the optional weighting unit 18 is eliminated, and to the output signal of the coring unit 12 weighted with the factor v2, if the weighting unit 18 is present. The weighting value v2 lies between 0 and 1 and is preferably 0.5 when transfer functions H1(z), H2(z) corresponding to the above equations (1) and (2) are used.

The limiting unit 16 is preferably designed to limit the high-pass filtered signal hp(k) symmetrically to an upper limit value g or a lower limit value −g, this limit value g being adjustable from the outside via a control input of the limiting unit 16. At the output of the limiting unit 16 there is a signal ghp(k) that is subtracted from the delayed input signal s(k). The limiting unit has the effect that the high-pass signal values ghp(k) subtracted from the input signal s(k) have a maximum amplitude g. High-pass signals components in the input signal s(k), whose amplitudes lie within the limiting region, are thus entirely low-pass filtered by the arrangement, while signal components whose amplitudes are larger than the limiting region are only reduced in amplitude by a value dictated by the limiting region. In particular, noise influences with low amplitude can be filtered out in this way, although useful high-frequency signal components will also be influenced.

FIG. 11 shows a modification of the low-pass filter arrangement in FIG. 10, which differs from the filter arrangement in FIG. 10 in that the limited high-pass filtered signal ghp(k) is subtracted by means of a subtractor 30 from the high-pass filtered signal hp(k), in order to generate a difference signal df(k). This difference signal df(k) is furnished by means of an additional limiter 34, which likewise receives the limit value g, and which likewise symmetrically limits the signal furnished to it. The signal ghp2(k) present at the output of this limiter 34 is added by means of an additional adder 36 to the output signal tp1(k) of the subtractor 14, which subtracts the limited high-pass filtered signal ghp(k) from the delayed input signal s(k), in order to furnish a low-pass filtered signal tp2(k). As was mentioned in connection with FIG. 10, high-frequency signal components which are smaller than the limiting value are subtracted from the input signal s(k) and thus low-pass filtered, but useful high-frequency signal components will also be reduced in their amplitude. This influencing of the useful signal is undesirable in applications whose objective is to filter out only high-frequency noise signal components down to a particular amplitude.

In the arrangement in FIG. 11, for high-frequency signal components that are larger than the limiting range, a correction signal is added by means of the adder 36 to the low-pass filtered signal. For this, the difference signal df(k) is first produced, this signal df(k) being zero as long as the amplitude of the signal hp(k) is smaller than the limiting value g. For amplitudes of hp(k) larger than the limiting value, the amplitudes of the difference signal df(k) will be larger than zero. In order to avoid adding high-frequency signal components whose amplitudes are larger than the amplitudes of these signal components in the input signal s(k) to the signal tp1(k) obtained by subtraction of the input signal s(k) and the high-pass signal hp(k), the amplitude of the difference signal is limited by the limiting device 34 according to the signal hp(k) in unit 16. The result is an overall behavior of the circuit which can be described as follows:

• • low-pass filtering is done for high-frequency signal components smaller than the limiting range,
  • no change occurs for high-frequency signal components larger than twice the limiting range,
  • for high-frequency signal components larger than the limiting range and smaller than twice the limiting range, there is a gradual transition from low-pass filtering to an uninfluenced passage of the signal.

FIG. 12 shows a low-pass filter arrangement modified from FIG. 11, which differs from that depicted in FIG. 11 in that a filter 32 for filtering the difference signal df(k) is connected ahead of the limiting unit 34, and the transfer behavior of this filter can be adjusted by means of an external adjustment value f. In this exemplary embodiment, the output of the limiting unit 34 presents a high-pass filtered signal ghp2′(k), which is added to the output signal tp1(k) of the subtractor. The output of the overall arrangement furnishes a low-pass filtered signal tp2′(k).

Filter 32 makes it possible to organize the previously mentioned transition range between a low-pass filtering of the high-frequency signal components and their unchanged passage. This transition region can become narrower if filter 32 has an amplifying nature with a gain factor larger than one, since limiting will then occur already for small values of the difference signal df(k).

This transition region can also become broader if filter 32 has an amplifying nature with a gain factor less than one, since then limiting will occur only for large values of the difference signal df(k).

Preferably, filter 32 has a transfer behavior that damps small values of the difference signal df(k), amplifies medium values, and again damps large values, preferably values that are only somewhat smaller than the limiting value, in order to organize a smooth transition between the low-pass behavior and the free passage behavior. The filter can contain a reference table whereby the individual gain factors are assigned to the amplitudes of the difference signal df(k).

FIG. 13 shows a modification of the filter arrangement in FIG. 12, wherein this low-pass filter arrangement differs from that depicted in FIG. 12 in that, besides the filtering with the filter 32 and the limiting 34, the difference signal df(k) is subjected to weighting with an adjustable weighting factor v3 by means of a weighting unit 38, wherein a signal present at the output of the weighting unit 38 is added by means of an adder 40 to the output signal tp2′(k) of the adder 36.

As was explained with reference to FIGS. 6 to 8, low-pass filtering with filter arrangements according to FIGS. 6 and 7—and also according to FIGS. 10 to 13—has the drawback that pulses of the input signal s(k) are damped in their amplitude depending on a weighting factor whereby the high-pass filtered signal hp(k) is weighted before subtraction from the input signal.

As represented in FIG. 14, this problem can be eliminated by connecting the first filter arrangement with the high-pass filter HP1 according to the invention to another filter arrangement downstream, which likewise has a high-pass filter according to the invention, preferably configured identical to the high-pass filter HP1 of the first filter arrangement. The second filter arrangement, whose input receives the output signal tp(k) of the first filter arrangement, differs from the first filter arrangement in that the signal present at the output of the high-pass filter is added to the input signal, in order to provide a filter output signal tp4(k) in the exemplary embodiment. While the first filter arrangement cuts in half the amplitude of pulses of the input signal s(k), as is seen by comparing the input signal s(k) and the signal tp(k) present at the output of the first filter arrangement in FIGS. 14b and 14c, these pulses are again filtered out in the second filter arrangement by means of the high-pass filter arrangement, yet added to the input signal, so as to restore once again the pulses with their original amplitude in the resulting filter signal tp4(k).

The improvement in the transfer behavior of the system according to FIG. 14 with the low-pass filter arrangement, which filters out the high-frequency signal components, and the following filter arrangements, which amplifies high-frequency signal components contained in the output signal of the first filter arrangement, as compared to a traditional system of this kind, is made clear by FIG. 16. FIG. 16a shows a prior art system constructed in accordance with the system in FIG. 14, where traditional high-pass filters are used instead of the high-pass filter arrangements according to the invention. These high-pass filters are designated in FIG. 16a by the reference number 200. Adders of this system, which correspond to the adders 14, 114 of the system according to FIG. 14a, are designated in FIG. 16a with the reference numbers 210, 220.

The system according to FIG. 16 receives an input signal in(k), which corresponds to the signal s(k) in FIG. 14b. For the transfer function of the high-pass filter 200, it is assumed that this transfer function corresponds to the transfer function of the high-pass filter 10 according to FIG. 14a. A comparison of the output signals out(k) in the prior art system and tp4(k) in the system according to the invention makes it clear that edges in the signal diagram remain fully intact in the system according to the invention, whereas they are flattened out in the known system, in addition to which pulses as well fully remain fully intact in the system with the high-pass filter arrangements according to the invention, whereas they are reduced in amplitude in the prior art system.

To summarize, it may be said that the high-pass filter method according to the invention and a high-pass filter arrangement according to the invention with an adaptively adjusted coring unit has diversified applications and that the high-pass filter arrangement according to the invention has advantages over traditional linear high-pass filter arrangements, especially when filtering step signals.

For the discussion of the method according to the invention it has been assumed thus far that the signal values processed came from a signal sequence with signal values consecutive in time or space, generating an output value for every input value by using neighboring values. Depending on the specific configuration of the method, the signal sequence furnished at the output represents a high-pass filtered or low-pass filtered input sequence.

However, the method according to the invention is not limited to the processing of signal sequences with consecutive signal values in time or space, but instead can also be applied to otherwise generated signal values, as will be explained hereinbelow, where the first signal value S1 replaces the former first signal value s(i), the second signal value S2 the former second signal value s(i−1), and the third signal value S3 the former third signal value s(i+1).

FIG. 15 shows a modification of the device according to FIG. 7. The device according to FIG. 7 serves to apply the method according to the invention to an input sequence s(k) of consecutive signal values, wherein the filters 10, 20 contain delay elements 107, 108, 207, 208, in order to process three consecutive signal values and furnish from these three consecutive signal values a filter output value hp(k) or tp(k).

In the device according to FIG. 15, these delay elements according to FIG. 7 are eliminated and instead a first signal value S1, a second signal value S2, and a third signal value S3 are furnished in parallel to the device.

Based on the fact that, in FIG. 7, an instantaneous input value s(k), an input value delayed by 1 or s(k−1), and an input value delayed by 2 or s(k−2), is processed at the same time by the two filters 10, 20, the processing of the first signal value S1 corresponds to the processing of the signal value s(k−1) in FIG. 7, the processing of the second signal value S2 corresponds to the processing of the signal value s(k), and the processing of the third signal value S3 corresponds to the processing of the signal value s(k−2).

The device according to FIG. 15 determines, from these three values S1, S2, S3, an output signal hp(k) at the output of the coring unit 12 or a filter output value tp(k) at the output of the subtractor 14. The filter output value hp(k) in the device in FIG. 15 corresponds to a measure of the mutual distance of signal values S1, S2, S3. The filter value tp(k), which subtracts the output value hp(k) of the coring unit from the first signal value S1, corresponds to an interpolation value generated from the first, the second, and the third signal value S1, S2, S3.

This device according to FIG. 15 or the underlying method, which furnishes an interpolated value tp(k) from three signal values S1, S2, S3, is especially suitable for pixel interpolation, as will be explained hereinbelow with reference to FIGS. 16 and 17.

FIG. 16 shows schematically two consecutive interlaced fields A, B of a field sequence, wherein field A, for example, comprises the even-numbered lines and field B the odd-numbered lines of the image. In order to interpolate a video information value for the pixel at position (x0, y0), in field A, one will use the video information values of three pixels, which are designated in FIG. 16 as P_A(x0−1, y0), P_A(x0+1, y0) and P_B(x0, y0). The first signal value S1 required for the method according to the invention corresponds here to the video information value P_B(x0, y0) at position (x0, y0) in the second field B. The second and third signal values S2, S3 come from the first field A and are the pixels neighboring the pixel (x0, y0) being interpolated in the vertical direction: the second video information value S2 corresponds to the value of the pixel P_A(x0−1, y0) lying above the pixel being interpolated, and the third pixel S3 corresponds to the value of the pixel P_A(x0−1, y0) lying below the pixel being interpolated.

If one furnishes these three signal values S1, S2, S3 to the device of FIG. 15, or applies the previously explained low-pass filter method to these three signal values, the signal value tp(k) furnished at the output for the three signal values S1, S2, S3 corresponds to the video information value P_A(x0, y0) being interpolated.

FIG. 17 shows schematically two consecutive images C, D, for which one is supposed to interpolate an intermediate image CD lying between these two images in time or space, by using motion vectors. (x0, y0) in the image CD being interpolated is the position of one of the pixels being interpolated. This pixel is assigned two motion vectors vec1, vec2, whose spatial orientation is represented in images C, D. Pixel x0, y0 lies on these motion vectors. In order to interpolate the pixel P_CD(x0, y0), one will use the video information values P_C(x1, y1), P_C(x3, y3) at the starting points (x1, y1) or (x3, y3) of the first and second motion vector vec1, vec2 in the first image C and the video information values P_D(x2, y2) and P_D(x4, y4) at the end points (x2, y2) and (x4, y4) at the end points of the first and second motion vector vec1, vec2 in the second image D. The first signal value S1 to be processed according to the method according to the invention corresponds here to a mixed value of the video information value at the starting point of the first vector vec1 in the first image C and at the end point of this first vector vec1 in the second image D, while the mixed value is preferably the mean value, so that we have: S1=(P_C(x1, y1)+P_D(x2, y2))/2

The second signal value S2 being processed corresponds to the video information value P_C(x3, y3) at the starting point of the second vector vec2 in the first image C and the third signal value S3 being processed corresponds to the video information value P_D(x4, y4) at the end point (x4, y4) of the second vector vec2 in the second image D.

If one furnishes these signal values S1, S2, S3 to the device of FIG. 15 or applies the previously explained low-pass filter method to these signal values, the signal value produced tp(k) corresponds to the video information value P_CD(x0, y0) being interpolated.

Although the application of the method according to the invention to pixel interpolation is explained by means of a modification of the device according to FIG. 7, it is pointed out that any of the low-pass filter methods explained in the present application are of course applicable to pixel interpolation, provided the at least three processed signal values are furnished in the manner explained with reference to FIGS. 16 and 17.

Claims

1. Method for high-pass filtering of a signal (s(k)) having a sequence of signal values, in order to furnish a high-pass filtered signal (hp(k)) having a sequence of signal values, wherein the method for producing a signal value (hp(i)) of the filtered signal (hp(k)) comprises the following steps: a) selection of a first signal value (s(i)) of the signal (s(k)) to be filtered and at least one second signal value (s(i−1)) preceding the first signal value (s(i)) in the signal sequence (s(k)) and at least one third signal value (s(i+1)) following the first signal value (s(i)) in the signal sequence (s(k)), b) determination of an interval dictated by the at least one second signal value (s(i−1)) and the at least one third signal value (s(i+1)), with a first interval boundary depending on the at least one second signal value (s(i−1)) and a second interval boundary dictated by the at least one third signal value (s(i+1)), c) determination of a mean value (m) from the at least one selected second signal value (s(i−1)) and the at least one selected third signal value (s(i+1)), d) determination as to whether the first signal value (s(i)) lies within the interval, e) if the first signal value (s(i)) lies outside the interval, determining a first mathematical distance (d1) between the first signal value (s(i)) and the first interval boundary (s(i−1)) and a second mathematical distance (d2) between the first signal value (s(i)) and the second interval boundary (s(i+1)) and a third mathematical distance between (d3) between the first signal value (s(i)) and the mean value (m) in order to furnish a first, a second, and a third distance value, selection of a value from a value range between the smaller of the first and second distance values (d1, d2) and the third distance value (d3), f) if the first signal value (s(i)) lies within the interval, selecting value from a value range between zero and half the interval width, g) choosing the sign of the selected signal value depending on the position of the first signal value (s(i)) in relation to the mean value (m) to furnish the filtered signal value (hp(k)).

2. Method according to claim 1, in which the value selected in steps e) or f) is weighted.

3. Method according to claim 1, in which the smaller of the first and second distance values (d1, d2) is selected as the selected value in step e), and zero is selected in step f).

4. Method according to claim 1, in which the sign of the signal value selected in steps e) or f) is made plus in step g) if the first signal value (s(i)) is larger than the mean value (m) or minus if the first signal value (s(i)) is smaller than the mean value (m).

5. Method according to claim 1, in which the distance value (d1, d2, d3) when determining the first, second, and/or third distance respectively is set at zero if the distance value is smaller than a given limit value.

6. Method according to claim 1, in which the filtered signal value hp(k) is limited to an upper limit value (g) and a lower limit value (−g).

7. Method according to claim 1, in which more than one second signal value (s(i−1)) and/or more than one third signal value (s(i+1)) is provided, wherein one of the second signal values is selected as the second signal value (s(i−1)) and one of the third signal values as the third signal value (s(i+1)).

8. Method according to claim 7, in which a mathematical distance is determined between each of the second signal values (s(i−1)) and each of the third signal values and the second signal value and the third signal value having the greatest distance or the smallest distance are selected as the first and second interval boundaries.

9. The method of claim 1, comprising: furnishing a high-pass filtered signal (hp(k)) from an input signal (s(k)) to be filtered, generating the low-pass filtered signal (tp(k), tp2(k), tp3(k)) from the high-pass filtered signal (hp(k)) and the input signal s(k)).

10. Method according to claim 9, in which the generation of the low-pass filtered signal (tp(k)) comprises subtraction of the high-pass filtered signal (hp(k)) from the input signal (s(k)).

11. Method according to claim 10, in which the high-pass filtered signal (hp(k)) is amplified or damped prior to the subtraction.

12. Method according to claim 9, in which the generation of the low-pass filtered signal (tp2(k)) comprises the following steps: limiting the high-pass filtered signal (hp(k)) to generate a limited high-pass filtered signal (ghp(k)), subtraction of the limited high-pass filtered signal (ghp(k)) from the input signal (s(k)) to generate a first low-pass filtered signal (tp1(k)), subtraction of the limited high-pass filtered signal (ghp(k)) from the high-pass filtered signal (hp(k)) to generate a difference signal (df(k)), limiting a signal depending on the difference signal (df(k)), adding the first low-pass filtered signal (tp1(k)) and the limited difference signal (df(k)) to form the low-pass filtered signal (tp2(k)).

13. Method according to claim 10, in which the difference signal (df(k)) is filtered prior to the limiting.

14. Method according to claim 13, in which the filtering comprises amplification of the difference signal (df(k)) with a gain factor depending on the particular amplitude of the difference signal (df(k)).

15. Method according to claim 13, in which the difference signal (df(k)) is weighted and the weighted difference signal is added to the first low-pass filtered signal (tp1(k)) and the limited difference signal (df(k)) in order to form the low-pass filtered signal (tp3(k)).

16. The method of claim 9, comprising: filtering the low-pass filtered signal by means of a method according to one of claims 1 through 8 to furnish a high-pass filtered signal (hp2(k)), adding the high-pass filtered signal (hp2(k)) to the low-pass filtered signal.

17. High-pass filter arrangement (1), for filtering an input signal (s(k)) having a sequence of signal values: having an input for furnishing the input signal (s(k)) and an output for producing a filter output signal (hp(k)), a first high-pass filter (10, 20) and second high-pass filter, each of them connected behind the input, a coring unit (12) with a signal input, which receives a signal (hp1(k)) depending on a filter output signal of the first high-pass filter (10), with a control input, which receives a signal p(k)) dependent on a signal at the output of the second high-pass filter (20), and with an output, which is coupled to the output of the high-pass filter arrangement (1).

18. Filter arrangement according to claim 17, in which an arrangement (22) for forming the absolute magnitude is connected between the output of the second high-pass filter (20) and the control output of the coring unit (12).

19. Filter arrangement according to claim 17, in which an amplifier arrangement (24) is connected between the output of the second high-pass filter (20) and the control input of the coring unit (12).

20. Filter arrangement according to claim 16, in which the amplifier arrangement has a gain between 0 and 1.

21. Filter arrangement according to claim 15, in which the first high-pass filter (10) has a transfer function H1(z), for which:

22. Filter arrangement according to claim 21, in which the constant k1 is equal to 0.5 or equal to 0.25.

23. Filter arrangement according to claims 17, in which the second high-pass filter has a transfer function H2(z), for which:

24. Filter arrangement according to claim 23, in which the constant k2 is equal to 0.5.

25. Filter arrangement according to claim 17, in which the coring unit (12) has a transfer behavior for which:

26. Filter arrangement according to claim 17, in which a limiting arrangement (16) is connected between the output of the coring unit (12) and the output of the filter arrangement.

27. (canceled)

28. Method for generating a filter signal value, comprising the following steps: a) furnishing a first signal value S1, a second signal value (S2), and a third signal value (S3), b) determining an interval dictated by the at least one second signal value (S2) and the at least one third signal value (S3), with a first interval boundary depending on the at least one second signal value (S2) and a second interval boundary dictated by the at least one third signal value (S3), c) determining a mean value (m) from the at least one selected second signal value (S2) and the at least one selected third signal value (S3), d) determining whether the first signal value (S1) lies within the interval, e) if the first signal value (S1) lies outside the interval, determining a first mathematical distance (d1) between the first signal value (S1) and the first interval boundary (S2) and a second mathematical distance (d2) between the first signal value (S1) and the second interval boundary (S3) and a third mathematical distance (d3) between the first signal value (S1) and the mean value (m) to furnish a first, a second, and a third distance value, selecting a value from a value range between the smaller of the first and second distance values (d1, d2) and the third distance value (d3), f) if the first signal value (S1) lies within the interval, selecting a value from a value range between zero and the half-interval width, g) choosing the sign of the selected signal value depending on the position of the first signal value (S1) relative to the mean value (m) to furnish the filtered signal value (hp(k)).

29. Method according to claim 28, in which the filtered signal value is subtracted from the first signal value (S1) to furnish a filtered signal value (tp(k)).

30. Use of a method according to claim 29 for pixel interpolation when generating a frame from first and second consecutive interlaced fields, wherein the first signal value (S1) corresponds to a video information value of the second field (B) at the position (x0, y0) of the pixel being interpolated, the second signal value (S2) corresponds to a video information value of the first field (A) at a first position (x0−1, y0) neighboring the position (x0, y0) of the pixel being interpolated, the third signal value (S3) corresponds to a video information value of the first field (A) at a second position (x0+1, y0) neighboring the position (x0, y0) of the pixel being interpolated.

31. Use of a method according to claim 29 for vector-based interpolation of a pixel of an intermediate image, which lies in time or space between a first and second image (C, D), wherein the pixel is assigned at least a first and a second motion vector (vec1, vec2) and wherein the first signal value (S1) corresponds to a mixed value from the video information value of the pixel at the starting point (x1, y1) of the first vector (vec1) in the first image (C) and the pixel at the end point (x2, y2) of the first vector (vec1) in the second image (D), the second signal value (S2) corresponds to the video information value of the pixel at the starting point (x3, y3) of the second vector in the first image (C), the third signal value (S3) corresponds to the video information value of the pixel at the end point (x4, y4) of the second vector (vec2) in the second image (D).

32. Method according to claim 31, in which the mixed value corresponds to the mean value of the pixels at the starting point and end point of the first vector (vec1, vec2) in the first and second image (C, D).