The present invention relates to a noise elimination apparatus and noise elimination method for removing noise of a video signal.
A conventional noise elimination apparatus attenuates the interframe difference of a video signal at a ratio determined in accordance with conditions, and removes noise contained in the difference values. For example, in Patent Document 1, a motion deciding unit detects motion according to values based on the difference values to determine a ratio K (0<K<1), and obtains pixel values by adding previous frame pixel values multiplied by the ratio K and input pixel values multiplied by the ratio (1−K), thereby obtaining pixel values that attenuate noise contained in the interframe differences. Such a processing section is referred to as a noise elimination filter.
For example, as shown in the Patent Document 1, the conventional noise elimination apparatus applies the input pixel values and the previous frame pixel values to noise elimination calculation processing based on a ratio of (1−K):K. In other words, assume that the frame number proceeds with the time elapsed, and that the input pixel value (input image) of a given frame number t in the noise elimination processing is ft, and the pixel value after the noise elimination is Ft, then Ft is given by the following expression (1). Here, the frame number t is assumed to be an integer, and Ft-1 designates a pixel value (reference image) of the previous frame after the noise elimination.
Ft=(1−K)·ft+K·Ft-1 (1)
Although not described in the Patent Document 1, the video signal is generally handled in terms of integers. Accordingly, assuming that int(x) is an integerization function for integerizing a real number x, the integerized pixel value Ft after the noise elimination is given by the following expression (2).
Ft=int((1−K)·ft+K·Ft-1) (2)
In this case, since both Ft-1 and Ft are integers, the foregoing expression (2) can be rewritten as the following expression (3) which is expressed as the sum of the integerized pixel value Ft-1 after the noise elimination of the previous frame and the value (referred to as “output difference value”) obtained by multiplying the difference value (ft−Ft-1) between the input pixel value and the previous frame pixel value by the ratio (1−K).
Ft=Ft-1+int((1−K)·(ft−Ft-1)) (3)
For example, in
The integerization function int(x) includes integerization based on the rounding down processing and rounding up processing, and
First, removal of minute noise will be described in the case where the integerization is carried out by the rounding down processing and rounding up processing.
With respect to
When the difference values (ft−ft-1) between the corresponding input pixel values of consecutive frames take place with the percentages as shown in
Next, with respect to
When the difference values (ft−ft-1) between the corresponding input pixel values of consecutive frames take place with the percentages as shown in
The foregoing is the description about the effect on the removal of minute noise due to the difference between the rounding down processing and the rounding up processing in the integerization.
Next, the effect on the removal of minute noise due to the difference between the rounding down processing and the rounding up processing in the integerization will be described in the case where a screen gradually makes a transition from a certain still image to another different still image because of transition such as a dissolve in a conventional noise elimination apparatus. It is assumed here that (1−K) in the integerization function in the foregoing expression (3) is 1/4 when the absolute value of the difference value (ft−Ft-1) is equal to or less than four, and is one when it exceeds four as in
First, a case that applies the rounding down processing shown in
In the graph at the bottom of
At t=12, the transition of the input pixel values ft ends. The difference values with magnitude 4 up to t=12 become magnitude 3 at t=13 at which they are changed to reduce. From that point on, although the difference values are considered to be minute noise and are eliminated, the output pixel values Ft never exceed five and end the transition with being maintained at five, which is perceived as afterimages.
Next, a case that applies the rounding up processing shown in
In the graph at the bottom of
Furthermore, a case that applies the integerization function when the difference between the input pixel values and the output pixel values of the previous frame is equal to or greater than two will be described. In the case of applying the rounding down processing shown in
Likewise, in the case where the rounding up processing shown in
Here, the description is made by way of example of the dissolve that makes a linear transition with the difference of constant magnitude. However, a similar phenomenon can be confirmed as to the minute noise even in the nonlinear transition with the difference of variable magnitude.
As described above, when applying as the integerization function the rounding down processing to the calculation processing of the noise elimination, although the conventional noise elimination apparatus can eliminate minute noise, afterimages are sometimes perceived. On the other hand, when applying the rounding up processing as the integerization function, although it can prevent afterimages from being perceived, it cannot sometimes eliminate the minute noise.
Patent Document 1: Japanese Patent Laid-Open No. 6-225178/1994 (Paragraph 0013, FIGS. 1 and 6).
With the foregoing configuration, the conventional noise elimination apparatus has a problem of being unable to achieve the removal of minute noise and the elimination of afterimages at the same time.
The present invention is implemented to solve the foregoing problem. Therefore it is an object of the present invention to provide a noise elimination apparatus and noise elimination method capable of removing the minute noise and eliminating afterimages at the same time.
A noise elimination apparatus in accordance with the present invention includes: a subtracting section for obtaining a difference value between input image data and reference image data; a calculation processing section for carrying out calculation processing of multiplying the difference value obtained by the subtracting section by a prescribed coefficient; an adaptive integerization processing section for obtaining an integerized difference value by adaptively performing rounding up processing or rounding down processing on the difference value having undergone the calculation processing by the calculation processing section; an adding section for obtaining output image data on the basis of the integerized difference value obtained by the adaptive integerization processing section and the reference image data; and a memory for storing the output image data obtained by the adding section as the reference image data.
The noise elimination apparatus in accordance with the present invention has an advantage of being able to remove minute noise and eliminate afterimages at the same time.
The best mode for carrying out the invention will now be described with reference to the accompanying drawings to explain the present invention in more detail.
In the embodiment 1 in accordance with the present invention, a noise elimination apparatus will be described which includes an adaptive integerization processing section for adaptively selecting and applying one of the rounding down processing and rounding up processing as the integerization function for real values resulting from the calculation processing of the noise elimination.
The adaptive integerization processing section 13a includes, as a more detailed internal configuration, an integerization control section 21 having a frame counter 31, a period setting section 32, and a rounding up/down deciding section 33; and a rounding up/down processing section 22.
Referring to
The subtracting section 11 takes differences, at the corresponding positions, between image data (ft) 101 of a frame which is an input image and image data (Ft-1) 102 of a first previous frame which is stored in the frame memory 15 as a reference image, and outputs (ft−Ft-1) as difference values (d) 103 including noise components.
The calculation processing section 12a outputs a difference value (d′) 104 having undergone the noise elimination calculation processing by multiplying the difference value (d) 103 output from the subtracting section 11 by a prescribed coefficient α corresponding to (1−K) of the foregoing expression (3) described in the background art. Although the calculation processing section 12a can employ various types of noise elimination calculation processing, it is assumed here that it carries out calculation having nonlinear characteristics, and the description will be made by applying the characteristics as shown in
First, when the difference value (d) 103 is a target value of the noise elimination, that is, when the absolute value of the difference value (d) 103 is equal to or less than the threshold 4 in accordance with the threshold 4 in the nonlinear characteristics, the calculation processing section 12a obtains the difference value (d′) 104 having undergone the calculation processing of multiplying the prescribed coefficient α as shown in the following expression (4).
d′=α×d (4)
In this way, the calculation processing section 12a multiplies the difference value (d) 103 which is a noise component to be eliminated by the prescribed coefficient α less than one (0<α<1), 1/4 in this example, thereby carrying out the noise elimination calculation processing.
In contrast, when the difference value (d) 103 is a value outside the noise elimination target, that is, when the absolute value of the difference value (d) 103 is greater than the threshold 4 in accordance with the threshold 4, the calculation processing section 12a obtains the difference value (d′) 104 having undergone the calculation processing that multiplies the difference value (d) 103 by the prescribed coefficient α=1 in accordance with the foregoing expression (4). In this case, however, since the noise elimination calculation processing is substantially unnecessary, it is also possible to output the integer difference value (d) 103 with considering it as the real number difference value (d′) 104.
As for the prescribed coefficient α and the threshold in the nonlinear characteristics, they can be set independently of each other.
The adaptive integerization processing section 13a performs the adaptive integerization which will be described below on the difference value (d′) 104 on which the calculation processing section 12a carries out the noise elimination calculation processing, and outputs an integerized difference value (int(d′)) 105.
The adding section 14 adds the integerized difference value (int(d′)) 105 output from the adaptive integerization processing section 13a and the reference image data (Ft-1) 102 of the first previous frame stored in the frame memory 15 to obtain output image data (Ft) 106 having undergone the noise elimination. The frame memory 15 stores the output image data (Ft) 106 having undergone the noise elimination.
In this way, the noise elimination apparatus with the adaptive integerization processing section 13a in the embodiment 1 in accordance with the present invention operates. Here, the detailed operation of the adaptive integerization processing section 13a will be described.
The adaptive integerization processing section 13a comprises the integerization control section 21 and the rounding up/down processing section 22. The rounding up/down processing section 22 carries out the adaptive integerization which will be described below on the difference value (d′) 104 which has undergone the noise elimination calculation processing and is input to the adaptive integerization processing section 13a, and outputs the integerized difference value (int(d′)) 105 as the output of the adaptive integerization processing section 13a.
First, the frame counter 31 of the integerization control section 21 is a counter for counting up frame by frame input, and for outputting a count value (t) 201 of the frame.
The period setting section 32 of the integerization control section 21 has a period (N) 202 set for carrying out the rounding up processing, and outputs it. The period (N) 202 set in the period setting section 32 can be set in advance or set from outside such as a CPU (not shown).
Incidentally, although the description will be made in the assumption that the count value (t) 201 of the frame counter 31 is the same as the frame number t, the count value 201 can be set independently of the frame number to perform counting.
The rounding up/down deciding section 33 of the integerization control section 21 receives as its input the count value (t) 201 of the frame counted by the frame counter 31 and the period (N) 202 for carrying out the rounding up processing output from the period setting section 32; and provides the rounding up/down processing section 22 with the instruction 203 on the integerization function to be applied so that it carries out the rounding up processing when the count value (t) 201 of the frame to be subjected to the processing is divisible by the period (N) 202 for carrying out the rounding up processing, whereas it carries out the rounding down processing for an indivisible frame.
According to the instruction 203 given by the rounding up/down deciding section 33, the rounding up/down processing section 22 carries out the adaptive integerization that applies the rounding up processing or rounding down processing to the difference value (d′) 104 having undergone the noise elimination calculation processing by the calculation processing section 12a, and outputs the integerized difference value (int(d′)) 105.
In this way, in the integerization control section 21, the rounding up/down deciding section 33 and the rounding up/down processing section 22 carry out the integerization processing as indicated by the following expression (5) on the basis of the relationships between the count value (t) 201 and period (N) 202 of the frame.
In accordance with the upper equation of the foregoing expression (5), the rounding up/down deciding section 33 instructs to select and apply the rounding up processing if a remainder of the division of the count value (t) 201 of the frame by the period (N) 202 is zero, and the rounding up/down processing section 22 instructed to carry out the rounding up processing selects the rounding up processing and performs the integerization processing. On the other hand, in accordance with the lower equation of the foregoing expression (5), the rounding up/down deciding section 33 instructs to select and apply the rounding down processing if a remainder of the division of the count value (t) 201 of the frame by the period (N) 202 is other than zero, and the rounding up/down processing section 22 instructed to carry out the rounding down processing selects the rounding down processing and performs the integerization processing.
Here, although the rounding up processing is applied when the remainder of the division of the count value (t) 201 by the period (N) 202 is zero in the foregoing expression (5) and
Next, the operation at the time when noise with magnitude 1 occurs in the image data at any given frame number t will be described with reference to
Since the period (N) 202 of the rounding up processing is four, the rounding up/down deciding section 33 outputs the instruction 203 for the rounding up processing at the percentage of 1/N, that is, 1/4. As the integerized difference value (d″) 105, the output of the rounding up/down processing section 22, one is output at a ratio 1/4 of applying the rounding up processing in the total integerization processing, and the output image Ft having undergone the noise elimination becomes (x+1), thereby noise remains behind. However, at the ratio (N−1)/N of applying the rounding down processing, that is, 3/4, zero is output as the integerized difference value (d″) 105, and the output image Ft having undergone the noise elimination becomes x, thereby being able to eliminate the noise. In this manner, when the period (N) 202 of the rounding up processing is four, the minute noise can be eliminated at the ratio of 3/4. The noise can be eliminated at a higher ratio with an increase in the period (N) 202 for carrying out the rounding up processing. For example, when the period (N) 202 is eight, the minute noise can be eliminated at a ratio of 7/8.
The foregoing is the description about the removal of the minute noise by the adaptive integerization in the embodiment 1 in accordance with the present invention.
Next, the effect of the adaptive integerization on the removal of minute noise will be described when shifting from a certain still image to another different still image in the transition such as a dissolve in the noise elimination apparatus of the embodiment 1 in accordance with the present invention. It is assumed here that the prescribed coefficient α in the integerization function of the foregoing expression (4) is 1/4 in the range where the absolute value of the difference value (ft−Ft-1) is equal to or less than four, and is one in the ranges where it exceeds four as in
As
When the adaptive integerization function of the present embodiment 1 in accordance with the present invention is applied, the pixel values Ft having undergone the noise elimination for the input pixel values ft make a transition as indicated by circles in the graph at the top of
Incidentally, thick broken lines in the graph at the top of
In the case where the difference value d=ft−Ft-1 is equal to or less than four, the prescribed coefficient α=1/4 in the noise elimination calculation processing is applied. When d is 1-3, the difference after the calculation processing becomes less than one. In other words, from t=5 to t=7, the difference values having undergone the noise elimination calculation processing are subjected to the integerization by the rounding down processing, thereby resulting in the integerized difference values of zero. Thus, according to the foregoing expression (3) that uses (1−K) as the prescribed coefficient α, Ft becomes Ft=Ft-1.
Next, in the range from t=8 to t=12, d takes a value d=4, and it is integerized by the rounding up processing at t=8 and t=12, and by the rounding down processing at the remaining values. In this case, all the integerized difference values become one, and Ft becomes Ft=Ft-1+1 according to the foregoing expression (3) as well.
In addition, in the range from t=13 to t=16, d takes a value d=3, and it is integerized by the rounding down processing up to t=15, in which case the integerized difference value becomes zero, and according to the foregoing expression (3), Ft becomes Ft=Ft-1. At t=16, it is integerized by the rounding up processing, in which case the integerized difference value becomes one, and according to the foregoing expression (3), Ft becomes Ft=Ft-1+1. Although the input pixel value ft completes the transition and takes a fixed value, the difference value between the input pixel value and the pixel value of the previous frame having undergone the noise elimination is reduced by one, thereby alleviating afterimages.
Likewise, in the range from t=17 to t=20 where d takes a value d=2 and in the range from t=21 to 24 where d takes a value d=1, it is integerized by the rounding up processing at t=20 and t=24, in which case the integerized difference value becomes one. Thus, the difference values between the input pixel values and the pixel values of the previous frame having undergone the noise elimination are reduced one by one, thereby eliminating afterimages at t=25 in the end.
In this way, as for the period N for applying the rounding up processing in which the integerization based on the rounding up processing is applied, it is possible to mention that the difference value that causes afterimages at the completion of the transition of the input pixel value ft is reduced one by one at each period (N).
In this case, the transition state alters because the input pixel values ft, which make transition due to a dissolve, translate in the horizontal direction and the applying positions of the integerization based on the rounding up processing change. However, unlike the noise elimination according to the integerization that applies only the rounding down processing as described in the background art, the present embodiment 1 can reach Ft=ft through the periodic rounding up processing as long as the difference value is present, thereby being able to eliminate afterimages. In addition, in spite of the expansion and contraction in the vertical axis, the afterimages can also be eliminated even through the convergence time up to the elimination of the afterimage will change.
As for the prescribed coefficient α of the noise elimination calculation processing and the period N for applying the rounding up processing in the integerization function, they can be set independently of each other. In addition, as for the threshold in the nonlinear characteristics shown in
As described above, the present embodiment 1 in accordance with the present invention has the adaptive integerization processing section 13a that adaptively applies, as the integerization function, one of the rounding down processing and rounding up processing to the difference values having undergone the noise elimination calculation processing. As a result, it offers an advantage of being able to remove the minute noise and to eliminate the afterimages at the same time.
In an embodiment 2 in accordance with the present invention, a noise elimination apparatus will be described which has an adaptive integerization processing section that stores in a fraction memory fractions rounded down in the integerization of the corresponding pixels of the previous frame, and applies the integerization after reflecting the fractions of the previous frame on the real values removed by the noise elimination.
The adaptive integerization processing section 13b includes, as a more detailed internal configuration, a stillness deciding section 41a, a fraction memory 42a, a real number adding section 43, and an integer part/fractional part separating section 44.
Referring to
As in the foregoing embodiment 1, the subtracting section 11 takes differences, at the corresponding positions, between image data (ft) 101 of a frame which is an input image and image data (Ft-1) 102 of a first previous frame which is stored in the frame memory 15 as a reference image, and outputs (ft−Ft-1) as difference values (d) 103 including noise components.
If the stillness deciding section 41a of the adaptive integerization processing section 13b which will be described later obtains a decision result 301 that the input image data 101 is a still image portion, the calculation processing section 12b multiplies the difference value (d) 103 output from the subtracting section 11 by the prescribed coefficient α(0<α<1) in the foregoing expression (4), 1/4 as in the foregoing embodiment 1 here, and outputs the difference value (d′) 104 having undergone the noise elimination calculation processing. Unless the stillness deciding section 41a obtains the decision result 301 that the input image data (ft) 101 is a still image portion, the calculation processing section 12b considers the difference value (d) 103 output from the subtracting section 11 as the difference value (d′) 104 having undergone the noise elimination calculation processing, and outputs it.
As for the difference value (d′) 104 output when the decision result 301 that the input image data (ft) 101 is not a still image portion is obtained, although it remains an integer substantially, it can also be interpreted that the difference value (d) 103 is multiplied by one as the prescribed coefficient α and is made a real value.
The adaptive integerization processing section 13b performs the adaptive integerization which will be described below on the difference value (d′) 104 having undergone the noise elimination calculation processing by the calculation processing section 12b, and outputs the integerized difference value 105.
The adding section 14 adds the integerized difference value 105 output from the adaptive integerization processing section 13b and the image data (Ft-1) 102 of the first previous frame stored in the frame memory 15 to obtain output image data (Ft) 106 having undergone the noise elimination. The frame memory 15 stores the output image data (Ft) 106 having undergone the noise elimination as in the foregoing embodiment 1.
In this way, the noise elimination apparatus with the adaptive integerization processing section 13b in the embodiment 2 in accordance with the present invention operates. Next, the detailed operation of the adaptive integerization processing section 13b will be described.
The adaptive integerization processing section 13b comprises the stillness deciding section 41a, the fraction memory 42a, the real number adding section 43, and the integer part/fractional part separating section 44; and the difference value (d) 103, which is the input to the adaptive integerization processing section 13b, is input to the stillness deciding section 41a, and the difference value (d′) 104 having undergone the noise elimination calculation processing is input to the real number adding section 43. In addition, the integerized difference value 105, which is the output from the adaptive integerization processing section 13b, is output from the integer part/fractional part separating section 44.
First, the stillness deciding section 41a makes a decision from the difference value (d) 103 output from the subtracting section 11 as to whether the input image data (ft) 101 is a still image portion or not, and outputs the decision result 301. In this case, the stillness deciding section 41a makes a decision, for example, that the input image data 101 is a still image portion when the absolute value of the difference value (d) 103 is less than a threshold, and that the input image data 101 is not a still image portion when the absolute value is equal to or greater than the threshold. The decision that it is not a still image portion is considered to be equivalent to the decision that it is a moving image portion. The decision threshold is set at four, for example.
The fraction memory 42a, which is a memory for retaining, at respective pixel positions, fractions Ct-1 in decimal places which are rounded down at the time of integerization of the first previous frame in the integer part/fractional part separating section 44, retains the fractions in decimal places as long as the decision result 301 by the stillness deciding section 41a is a still image portion, and resets the fractions in decimal places to zero unless the decision result 301 is a still image portion, thereby causing to be referred to. Here, as for the accuracy of the fractions retained in the fraction memory 42a, if the prescribed coefficient α is a reciprocal of a power of two (assumed to be the Sth power), it is enough to retain the S digits of the numerical value representing the exponent part. For example, when the prescribed coefficient α is 1/4, it has two decimal places. In the case of the reciprocal of a power of two, the number of digits less than S digits is possible if the rounding down error due to cancellation of significant digits is allowable. In the case of not the reciprocal of a power of two, any appropriate number of digits can be used if the rounding down error of lower decimal places of the fractions retained is allowable. In the case of handling both the positive and negative difference values, it is assumed that one extra digit is retained to memorize the plus and minus sign.
The real number adding section 43 adds the difference value (d′) 104 having undergone the noise elimination calculation processing by the calculation processing section 12b and the fraction (Ct-1) 302 in decimal places of the first previous frame retained in the fraction memory 42a, and outputs the addition result as a real number difference value (d′+Ct-1) 303.
Here, the fraction (Ct-1) 302 added has been reset to zero before the addition unless the stillness deciding section 41a makes a decision that the input image data (ft) 101 is a still image portion.
The integer part/fractional part separating section 44 separates the real number difference value 303 into the integer part constituting an integerized difference value (d″) 105 and the fractions (Ct) 304 in decimal places, and outputs the integer part constituting the integerized difference value (d″) 105 to the adding section 14 and outputs the fraction (Ct) 304 in decimal places to the fraction memory 42a. The fraction (Ct) 304 in decimal places is retained in the fraction memory 42a to be used for calculating the real number difference value 303 at the same pixel position of the next frame.
When the stillness deciding section 41a does not make a decision that the input image data (ft) 101 is a still image portion, the difference value (d′) 104 having undergone the noise elimination calculation processing by the calculation processing section 12b is an integer difference value (d) 103 in practice, and the fraction (Ct-1) 302 retained in the fraction memory 42a is also reset before the addition by the real number adding section 43 so that the fraction (Ct) 304 retained in the fraction memory 42a becomes zero.
In this configuration, the operation in the case where noise with magnitude 1 occurs in the image data of any given frame number t will be described with reference to
The fraction memory 42a stores some fractions (Ct-1) 302 processed in the first previous frame, and when the prescribed coefficient α=1/4, they take one of the values given by Ct-1=0.00, 0.25, 0.50, and 0.75 at every 0.25 step. For the fractions (Ct-1) 302 from 0 to 0.50, since the real number difference value (d′+Ct-1) 303 output from the real number adding section 43 is less than one, the integerized difference value (d″) 105 consisting of the integer part output from the integer part/fractional part separating section 44 is zero, and the fraction (Ct) 304 consisting of the fractional part output takes one of the values 0.25, 0.50, and 0.75. Thus, it has the effect equivalent to the integerization based on the rounding down processing as the output of the adaptive integerization processing section 13b. In contrast, when Ct-1=0.75, the real number difference value (d′+Ct-1) 303 is one, the integerized difference value (d″) 105 of the integer part is one, and the fraction (Ct) 304 of the fractional part is 0.00. This brings about the effect equivalent to the integerization based on the rounding up processing as the output of the adaptive integerization processing section 13b. As described in the foregoing embodiment 1, the integerization corresponding to the rounding up processing cannot achieve the noise elimination. However, when the prescribed coefficient α=1/4, and if the output fraction (Ct) 304 of the fractional part of the integer part/fractional part separating section 44 and the fraction (Ct-1) 302 output from the fraction memory 42a, both of which take value at every 0.25 step, occur equally, the minute noise can be eliminated at a ratio of 3/4.
The foregoing is the description of the effect of the adaptive integerization on the removal of minute noise in the embodiment 2 in accordance with the present invention.
Next, the effect of the adaptive integerization on the removal of minute noise will be described when shifting from a certain still image to another different still image in the transition such as a dissolve in the noise elimination apparatus of the embodiment 2 in accordance with the present invention. It is assumed here that the prescribed coefficient α in the integerization function of the foregoing expression (4) is 1/4 when a decision of a still image portion is made from the difference value (ft−Ft-1), and is one unless a decision of a still image portion is made.
As
When the adaptive integerization function of the present embodiment 2 in accordance with the present invention is applied, the pixel values Ft having undergone the noise elimination for the input pixel values ft make a transition as indicated by circles in the graph at the top of
Incidentally, as in the embodiment 1 in accordance with the present invention, thick broken lines in the graph at the top of
When a decision is made that the difference value d in the middle graph is a still image portion, the real number difference value (d×α+Ct-1) denoted by a circle in the bottom graph takes a value equal to the square fraction Ct-1 plus (d/4) if the prescribed coefficient α for carrying out the noise elimination calculation processing is 1/4. The integer part of the real number difference value (d×α+Ct-1) before the integerization becomes the integerized difference value after the adaptive integerization. Thus, when the real number difference value exceeds one, the integerized difference value is one, and a portion exceeding one is the fraction Ct. In addition, when the real number difference value is less than one, the integerized difference value is zero, and the difference value itself is the fraction Ct. The fractions Ct of the current frame t obtained in this manner become the fractions of the next frame (t+1), which are denoted by squares. Accordingly, in the bottom graph, concerning the frame indicated by solid circles, the rounding up processing is carried out, and concerning the frame indicated by open circles, the rounding down processing is carried out, thereby producing the output pixel values Ft having undergone the noise elimination.
The present embodiment 2 in accordance with the present invention does not always result in the integerization based on the fixed period rounding down processing and rounding up processing as in the embodiment 1 in accordance with the present invention. However, the rounding up processing that occurs appropriately from t=7 can achieve Ft=ft at t=18, thereby being able to eliminate afterimages.
As described above, the present embodiment 2 in accordance with the present invention has the adaptive integerization processing section 13b that stores in the fraction memory 42a the fractions rounded down in the integerization of the corresponding pixels of the previous frame, and applies the integerization after reflecting the fractions on the real values removed by the noise elimination. Thus, the present embodiment 2 has an advantage of being able to remove the minute noise and to eliminate the afterimages at the same time.
In addition, the present embodiment 2 retains the rounded down fractions of the corresponding pixels of the previous frame in the fraction memory 42a, and carries out integerization by separating the integer part after adding the fractions to the real values having undergone the noise elimination by the calculation processing section 12b. Thus, even if the integerization function applied includes only the rounding down processing, the present embodiment 2 has an advantage equivalent to the integerization using the rounding up processing for the real values having undergone the noise elimination by the calculation processing section 12b.
In the embodiment 3 in accordance with the present invention, a noise elimination apparatus will be described in which the calculation processing section includes the functions corresponding to those of the stillness deciding section in the adaptive integerization processing section in the foregoing embodiment 2.
The adaptive integerization processing section 13c includes, as a more detailed internal configuration, a fraction memory 42a, a real number adding section 43, and an integer part/fractional part separating section 44; and is the same as that shown in
In
The operation of the fraction memory 42a that obtains the decision result 401 and the operation of the noise elimination apparatus from that point forward are the same as the description about the fraction memory 42a that obtains the decision result 301 in the foregoing embodiment 2.
As described above, the present embodiment 3 in accordance with the present invention has the same advantage as the foregoing embodiment 2, as the noise elimination apparatus with the configuration having the calculation processing section 12c including the stillness deciding section 41a located in the adaptive integerization processing section 13b of the foregoing embodiment 2.
In the present embodiment 4 in accordance with the present invention, the noise elimination apparatus will be described which makes independent the decision results the stillness deciding section of the foregoing embodiment 2 outputs and supplies to the calculation processing section and to the fraction memory.
The adaptive integerization processing section 13d includes, as a more detailed internal configuration, a stillness deciding section 41b, a fraction memory 42a, a real number adding section 43, and an integer part/fractional part separating section 44; and is the same as that shown in
In the foregoing embodiment 2, the stillness deciding section 41a makes a decision from the difference value 103 output from the subtracting section 11 as to whether the input image data 101 is a still image portion or not, and outputs the decision result 301. In the embodiment 4 in accordance with the present invention, the stillness deciding section 41b divides the reliability of the decision on the still image portion into two or more grades, notifies the calculation processing section 12b, and prepares a plurality of prescribed coefficients α for the noise elimination calculation processing in accordance with the classification. For example, the stillness deciding section 41b, which makes a decision from the difference value 103 output from the subtracting section 11, considers that the probability of not the still image portion becomes higher as the difference value increases. Thus, the reliability of making a decision that the input image data 101 is a still image portion decreases, and the effect of the noise on the image data reduces.
Accordingly, the difference value 103 is divided into a plurality of intervals using thresholds, and the prescribed coefficients α are increased to closer to one as the probability of not the still image portion increases using the intervals as the classification. For example, as the nonlinear characteristics for providing the prescribed coefficients, the intervals and the prescribed coefficients can be set rather freely such as: the prescribed coefficient α=1/4 in the interval from −2 to 2 inclusive, and α=1/2 in the intervals from −4 to −2 inclusive or from 2 to 4 inclusive, in which a decision is made that the difference value 103 corresponds to the still image portion; and α=1 in the remaining intervals in which a decision is made that the difference value 103 does not correspond to the still image portion.
As in the foregoing embodiment 2, when it makes the decision of not the still image portion as a decision result 301b, the stillness deciding section 41a notifies the fraction memory 42a to reset the fraction 302 via a route separate from the decision result 301a.
Thus, the noise elimination apparatus operates in the same manner as described in the foregoing embodiment 2 except that the stillness deciding section 41b supplies the calculation processing section 12b with the decision result 301a, and the fraction memory 42a with the decision result 301b; and that the calculation processing section 12b applies the prescribed coefficients α based on the decision result 301a.
In the embodiment 4 in accordance with the present invention, the decision results of the stillness deciding section supplied to the calculation processing section and fraction memory are described separately. However, enabling the fraction memory to distinguish a specified decision result to the calculation processing section from the other makes it possible to implement equivalent operation in
As described above, the embodiment 4 in accordance with the present invention offers the same advantages as the foregoing embodiment 2, as the noise elimination apparatus with the configuration of dividing the nonlinear characteristics of the calculation processing section 12b of the foregoing embodiment 2 into three or more states.
In addition, the present embodiment 4 in accordance with the present invention uses different prescribed coefficients for different divisions of the nonlinear characteristics defined by dividing the probability of making a decision of the still image portion into three or more states based on the difference value output from the subtracting section 11. Accordingly, the present embodiment 4 can assign a different prescribed coefficient to an intermediate portion between the still image portion and not the still image portion, can make it easier for the output of the real number adding section 43 to exceed one because of an increase in the fraction for the difference value, and can facilitate applying the integerization corresponding to the rounding up processing, thereby offering an advantage of being able to improve the delay.
In the present embodiment 5 in accordance with the present invention, a noise elimination apparatus will be described which improves the decision reliability by enabling the stillness deciding section of the foregoing embodiment 2 to refer to the decision result on neighboring image data at the time of making the decision.
The adaptive integerization processing section 13e includes, as a more detailed internal configuration, a stillness deciding section 41c, a fraction memory 42b, a real number adding section 43, and an integer part/fractional part separating section 44; and is the same as that shown in
The embodiment 5 in accordance with the present invention, utilizing the fact that the fraction memory 42b is reset to zero when the stillness deciding section 41c does not make a decision of the still image portion, makes a conjecture as to whether the input image data is a still image portion or not by referring to the retained fractions of the adjacent image data that have processed in the same frame, or of the adjacent image data that have not yet processed in the same frame, but correspond to those in the previous frame, thereby increasing the reliability of the decision. Referring to the fractions of the plurality of adjacent image data, the stillness deciding section 41c makes a conjecture that the input image data is not a still image portion if the fractions are zero. It is also possible to refer to the fraction at the same position in the previous frame before the reset based on the decision result is applied. For example, the stillness deciding section 41c makes a final decision whether the input image data is a still image portion or not according to the ratio of zeros therein and the magnitude of the difference value 103.
In addition, as in the foregoing embodiment 4, it is also possible to classify the decision reliability, and to make a decision on the classification by referring to the fractions of the adjacent image data.
As described above, the present embodiment 5 in accordance with the present invention can increase the reliability of the decision of the stillness deciding section 41a in the adaptive integerization processing section 13b of the foregoing embodiment 2, thereby being able to achieve an advantage similar to that of the foregoing embodiment 2.
The embodiments 1-5 in accordance with the present invention are described by way of example in which the minute noise is positive. However, it is obvious that if it is negative, the same advantage can be obtained by making a decision of the integerization by taking the absolute value of the negative minute noise. In addition, the minute noise is not limited to 1 or −1.
The embodiments 1-5 in accordance with the present invention are described by way of example that applies as the characteristics of the calculation processing section the example shown in
In addition, in the embodiments 1-5 in accordance with the present invention, according to a trend such as an increase, decrease, convergence and stagnation of the difference value in the calculation processing section, which is obtained from the secondary difference (derivative) value, the prescribed coefficients can be set separately within the same threshold interval.
As described above, the noise elimination apparatus in accordance with the present invention is suitably applied to such devices that have to prevent afterimages and to eliminate minute noise, for example.
Filing Document | Filing Date | Country | Kind | 371c Date |
---|---|---|---|---|
PCT/JP2006/307496 | 4/7/2006 | WO | 00 | 8/19/2008 |
Publishing Document | Publishing Date | Country | Kind |
---|---|---|---|
WO2007/116514 | 10/18/2007 | WO | A |
Number | Name | Date | Kind |
---|---|---|---|
4179710 | Ishiguro et al. | Dec 1979 | A |
4236224 | Chang | Nov 1980 | A |
4727506 | Fling | Feb 1988 | A |
4926361 | Ohtsubo et al. | May 1990 | A |
5487086 | Bhaskar | Jan 1996 | A |
5719793 | Nitta et al. | Feb 1998 | A |
6282243 | Kazui et al. | Aug 2001 | B1 |
6664913 | Craven et al. | Dec 2003 | B1 |
Number | Date | Country |
---|---|---|
2023365 | Dec 1979 | GB |
6-225178 | Aug 1994 | JP |
7-170429 | Jul 1995 | JP |
10-126688 | May 1998 | JP |
2000-22991 | Jan 2000 | JP |
2000-152036 | May 2000 | JP |
2002-94833 | Mar 2002 | JP |
2004-328206 | Nov 2004 | JP |
Number | Date | Country | |
---|---|---|---|
20090028457 A1 | Jan 2009 | US |