N-dimensional filter and method for n-dimensionally filtering an original image pixel

Abstract

The invention relates to known N-dimensional filters (100) and methods for N-dimensionally filtering for applying noise reduction to an original image pixel. However, carrying out only noise reduction to a pixel of an image is often not enough when a good picture quality is required. Thus, additionally sharpness enhancement (peaking) is carried out. However, noise reduction and peaking have contrasting spectral behaviors which leads to non-optimal results when both operations noise reduction and peaking are implemented in series. Starting from that prior art it is the object of the invention to provide noise reduction and sharpness enhancement to the original pixel in an easier and cheaper manner. This object is solved according to the present invention by defining coefficients α(p, n) or β(p, n) of said transmission function such that the transmission function is capable of applying noise reduction as well as sharpness enhancement to the input signal of said N-dimensional filter (100).

Description

[0001] The invention relates to an N-dimensional filter (100) and methods having a transmission function for applying noise reduction to an original image pixel being provided to the N-dimensional filter by a received input-signal, wherein said transmission function comprises at least one filter-coefficient.

[0002] The invention further relates to a method for N-dimensionally filtering an original image pixel p according to a transmission function in order to apply noise reduction to said pixel p.

[0003] Such N-dimensional filters or filtering methods are known in the art, e.g. from EP 0 682 841 B1. In said European patent specification a transmission function Ff(p) is disclosed which is defined as follows: 1$\begin{array}{cc}{F}_f\left(p\right)=G\left(p\right)·\left[F\left(p\right)+\gamma \sum _{n\in N_1}\alpha \left(p,n\right)·F\left(p+n\right)+\delta \sum _{n\in N_2}\beta \left(p,n\right)·F_f\left(p+n\right)\right]& \left(1\right)\end{array}$

[0004] wherein

[0005] N1, N2: are sets of vectors defining one, two or three dimensional neighbourhoods of one original image pixel;

[0006] n: identifies the position of an actual pixel within said neighbourhoods N1, N2;

[0007] G(p): is a normalization factor;

[0008] α(p, n),: are filter coefficients of said N-dimensional filter, in particular of said

[0009] β(p, n) transmission function Ff(p); and

[0010] γ, δ: are predefined constants.

[0011] Further, in said European patent specification the filter coefficients are defined as follows: 2$\begin{array}{cc}\alpha \left(p,n\right)=\{\begin{array}{c}\mathrm{w1}\text{:}\Delta \left(p,n\right)<{\mathrm{Th}}_1\\ \mathrm{w2}\text{:}{\mathrm{Th}}_1\le \Delta \left(p,n\right)>{\mathrm{Th}}_2\\ 0\text{:}\Delta \left(p,n\right)\ge {\mathrm{Th}}_2\end{array}& \left(2\right)\\ \beta \left(p,n\right)=\{\begin{array}{c}\mathrm{w1}\text{:}{\Delta }_f\left(p,n\right)<{\mathrm{Th}}_1\\ \mathrm{w2}\text{:}{\mathrm{Th}}_1\le {\Delta }_f\left(p,n\right)>{\mathrm{Th}}_2\\ 0\text{:}{\Delta }_f\left(p,n\right)\ge {\mathrm{Th}}_2\end{array}& \left(3\right)\end{array}$

[0012] wherein the parameters Δ and Δfare defined as follows:

Δ(p,n)=|F(p+n)−F(p)|  (4)

Δf(p,n)=|Ff(p+n)−F(p)|  (5)

[0013] and wherein

[0014] Th1, Th2: are predetermined threshold values; and

[0015] w1, w2: are positive integer values.

[0016] FIG. 4 shows an example for a typical distribution of the magnitudes of the filter coefficients α(p, n) or β(p, n).

[0017] It is important to note that according to EP 0 682 841 B1 the filter coefficients α and β are defined such that the transmission function carries out only noise reduction to the input signal, i.e. in particular to the original image pixel. In the case of the transmission function according to formula (1) that is achieved by defining the filter coefficients α and β always positive as done in formulas (2) and (3).

[0018] However, for improving the quality of images very often noise reduction is not enough. Usually also sharpness enhancement (peaking) of the image and in particular of the image pixels is additionally required in receivers of an image signal. Traditionally these two operations have been implemented in series. In these cases the noise filter has for example the transmission function (1) whereas the sharpness enhancement function is realised by another separated filter having quite another transmission function.

[0019] However, spectrally this serial arrangement does not produce an optimal result This is because noise reduction is commonly a low-pass-filtering operation, whereas peaking is a high-pass operation. Hence, there is a conflicting spectral demand on both filters and generally the optmisation of one leads to deterioration of the other. If the noise reduction is done after the peaking, then the noise filter will remove the sharpness enhancement created by a peaking filter. Usually, the pealing is done after the noise filtering as this leads to a more acceptable behaviour. However, this also requires some compromise since peaking attends to enhance remaining image noise.

[0020] Starting from that prior art it is the object of the invention to provide noise reduction and sharpness enhancement to the original pixel in an easier and cheaper manner.

[0021] Said object is solved for a known N-dimensional filter according to claim 1 in the way that at least one filter-coefficient of the transmission function is defined such that the N-dimensional filter is capable of applying noise reduction as well as sharpness enhancement to the signal input to the N-dimensional filter.

[0022] Both operations, noise reduction and sharpness enhancement, are mathematically combined and implemented within the transmission function of said one N-dimensional filter; as a result, when applying the transmission function to the signal, both operations are simultaneously applied to the signal.

[0023] Consequently, there is no need for applying a separate sharpness enhancement filter in series to a noise reduction filter. Therefore the solution can be implemented easily and cheap.

[0024] Advantageously when applying the transmission function to the original pixel each pixel in the neighbourhood of the original pixel contributes either to the noise reduction or to the sharpeness enhancement of the original pixel, but not to both.

[0025] According to a preferred embodiment the transmission function is defined according to claim 3. That transmission function Ff(p) has the advantage that it is very generic due to its recursive character. It enables the use of already filtered pixels for calculating the transmission function Ff(p) for new pixels.

[0026] For enabling the transmission function according to claim 2 to carry out noise reduction as well as sharpness enhancement the filter-coefficients α(p, n), β(p, n) of said transmission function are preferably designed as proposed in claim 4. There the coefficients are defined such that they either have a first or a second sign depending on as to whether a predetermined parameter Δ or Δf lies between predetermined threshold values Th3 and Th4 or not. The parameters Δ and Δf preferably represent a difference between the grey value of a neighbour pixel in said predetermined N-dimensional neighbourhood of the original pixel and the grey value of original pixel itself. The different signs of the filter coefficients ensure advantageously that the operations noise reduction and sharpness enhancement are realised by the same transmission function but are respectively applied to different neighbour pixels of the original pixel. In that mutually exclusive way the conflicting spectral behaviour of both operations is at least partly overcome.

[0027] It is advantageous to filter only a selected spectral component of the input signal by said N-dimensional filter. In that case the rest of said input signal, which is usually a video signal, is passed by said N-dimensional filter without filtering.

[0028] Further advantageous embodiments of the invention are subject matter of the remaining dependent claims.

[0029] The above identified object of the invention is further solved by the method described in claim 11. The advantages of said method correspond to the above described advantages of the N-dimensional filter.

[0030] Finally, the above object is solved by a display apparatus as claimed in claim 12.

[0031] There are four figures accompanying the description, wherein

[0032] FIG. 1 shows a two-dimensional window for spatial noise filtering according to the present invention;

[0033] FIG. 2 shows an example for the magnitudes of the filter coefficients α(p, n), β(p, n) according to the present invention;

[0034] FIG. 3 shows a display apparatus according to the invention; and

[0035] FIG. 4 shows an example for magnitudes of the filter coefficients α(p, n), β(p, n) according to the prior art.

[0036] In the following a preferred embodiment of the present invention will be described by referring to FIGS. 1 and 2.

[0037] The invention concerns preferably but not exclusively non-linear spatial recursive spatial or spatial-temporal filters. Consequently, the invention might e.g. be based on a non-linear recursive spatio-temporal filter having a transmission function Ff(p) as defined in equation 1. Said transmission function defines the filter output for an original image pixel input to said filter. Depending on as to whether one of the constants γ or δ is zero or not, the transmission function includes a different number of groups α, β of filter coefficients. In general for defining said filter output said transmission function carries out a mathematical combination of filter-coefficients α(p, n), β(p, n) which respectively belong to groups α and β of filter coefficients and which are—according to the present invention—defined as follows: 3$\begin{array}{cc}\alpha \left(p,n\right)=\{\begin{array}{c}\mathrm{w1}\text{:}\Delta \left(p,n\right)<{\mathrm{Th}}_1\\ \mathrm{w2}\text{:}{\mathrm{Th}}_1\le \Delta \left(p,n\right)>{\mathrm{Th}}_2\\ p\text{:}{\mathrm{Th}}_3\le \Delta \left(p,n\right)\le {\mathrm{Th}}_4\\ 0\text{:}\mathrm{otherwise}\end{array}& \left(6\right)\end{array}$4$\begin{array}{cc}\beta \left(p,n\right)=\{\begin{array}{c}\mathrm{w1}\text{:}{\Delta }_f\left(p,n\right)<{\mathrm{Th}}_1\\ \mathrm{w2}\text{:}{\mathrm{Th}}_1\le {\Delta }_f\left(p,n\right)<{\mathrm{Th}}_2\\ p\text{:}{\mathrm{Th}}_3\le {\Delta }_f\left(p,n\right)\le {\mathrm{Th}}_4\\ 0\text{:}\mathrm{otherwise}\end{array}& \left(7\right)\end{array}$

[0038] wherein the variables and parameters p, n, Δf, Th1-Th4, w1 and w2 are defined as described above. The threshold values Th1-Th4 are freely selectable.

[0039] However, the definition of coefficients α(p, n) and β(p, n) differs from their definitions in the prior art for the case that the parameters Δ or Δf lie respectively between Th3 and Th4; i.e. for the case that the difference between the grey value of a neighbour pixel and the grey value of the original pixel lies between the threshold values Th3 and Th4.

[0040] FIG. 1 shows a typical 2-dimensional arrangement of neighbour pixels surrounding an original, i.e. actually regarded, pixel of an image.

[0041] If the parameters Δ or Δf lie respectively between Th3 and Th4 the coefficients α(p, n) and β(p, n) have a negative sign and a magnitude of P≠0 which is freely programmable to a predetermined value. In that case peaking is allowed to be applied to all of those neighbour pixels which do not take part in weighted averaging, i.e. which are not subject matter of noise reduction. Noise reduction is applied to neighbour pixels the coefficients of which are positive. In this way filtering and peaking will be realised within the same filter in a mutually exclusive way when determining the filter output Ff(p) for the original pixel.

[0042] FIG. 2 shows an example for a typical distribution of the filter coefficients according to the present invention. It is important to note that according to the present invention these filter coefficients may not only be positive as indicated by w1 and w2 but also be negative as indicated by P. In that way the transmission function according to formula (1) is—according to the invention—enabled to apply not only noise reduction but also sharpness enhancenment to the original image pixel.

[0043] In an experimentally verified embodiment of a transversal spatial filter good results are achieved using the following neighbourhood N1: 5$N_1=\left\{\left[\begin{array}{c}0\\ 0\\ 0\end{array}\right]\left[\begin{array}{c}-2\\ 0\\ 0\end{array}\right]·\left[\begin{array}{c}-4\\ 0\\ 0\end{array}\right]·\left[\begin{array}{c}2\\ 0\\ 0\end{array}\right]·\left[\begin{array}{c}4\\ 0\\ 0\end{array}\right]·\left[\begin{array}{c}-2\\ -2\\ 0\end{array}\right]\right\}\left[\begin{array}{c}2\\ -2\\ 0\end{array}\right]\left[\begin{array}{c}6\\ -2\\ 0\end{array}\right]\left[\begin{array}{c}-6\\ -2\\ 0\end{array}\right]\left[\begin{array}{c}10\\ -2\\ 0\end{array}\right]\left[\begin{array}{c}0\\ -4\\ 0\end{array}\right]\left[\begin{array}{c}-4\\ -4\\ 0\end{array}\right]\left[\begin{array}{c}4\\ -4\\ 0\end{array}\right]\left[\begin{array}{c}8\\ -4\\ 0\end{array}\right]\left[\begin{array}{c}-8\\ -4\\ 0\end{array}\right]\left[\begin{array}{c}-2\\ 2\\ 0\end{array}\right]\left[\begin{array}{c}2\\ 2\\ 0\end{array}\right]\left[\begin{array}{c}6\\ 2\\ 0\end{array}\right]\left[\begin{array}{c}-6\\ 2\\ 0\end{array}\right]\left[\begin{array}{c}10\\ 2\\ 0\end{array}\right]\left[\begin{array}{c}0\\ 4\\ 0\end{array}\right]\left[\begin{array}{c}-4\\ 4\\ 0\end{array}\right]\left[\begin{array}{c}4\\ 4\\ 0\end{array}\right]\left[\begin{array}{c}8\\ 4\\ 0\end{array}\right]\left[\begin{array}{c}-8\\ 4\\ 0\end{array}\right]\}$

[0044] while selecting w1=1, w2=¼, Th2=4·Th1, Th3=5·Th1, Th4=8·Th1, γ=1, and β=0. Th1 is adapted to the noise level. P is programmable and has a value between −{fraction (1/16)} and 0.

[0045] It should be noted that several other configurations, e.g. on block, field or frame basis are possible.

[0046] FIG. 3 shows a display apparatus 200, e.g. a television set or a computer monitor, for displaying images. Said display apparatus comprising a N-dimensional filter 100 as claimed in claim 1 for processing input images before being displayed by said display apparatus.

Claims

1. N-dimensional filter (100) having a transmission function for applying noise reduction to an original image pixel p being provided to the N-dimensional filter (100) by a received input-signal; characterised in that at least one filter coefficient of the transmission function is defined such that the transmission function is capable of applying noise reduction as well as sharpness enhancement to the signal input to said N-dimensional filter (100).

2. The N-dimensional filter (100) according to claim 1, characterised in that each filter coefficient of said transmission function is respectively assigned to a different one of neighbour pixels being comprised in a N-dimensional neighbourhood of said original pixel and defines if the assigned neighbour pixel contributes either to noise reduction or to sharpeness enhancement of the original pixel.

3. The N-dimensional filter (100) according to claim 1, characterised in that the transmission function is defined as follows: 6Ff⁡(p)=G⁡(p)·[F⁡(p)+γ⁢∑n∈N1⁢α⁡(p,n)·F⁡(p+n)+δ⁢∑n∈N2⁢β⁡(p,n)·Ff⁡(p+n)] wherein p(x, y, t): represents the spatial and temporal position of the original or a neighbour pixel; N1, N2: are vectors defining a one, two or three dimensional neighbourhood of the original pixel; n: represents the position of a current neighbour pixel in a neighbourhood N1, N2; G(p): is a normalisation factor; α, β: are groups of filter coefficients α(p, n) and β(p, n) of the transmission function; γ, δ: are constants; and F(p): is an input luminance signal input to said N-dimensional filter (100).

4. The N-dimensional filter (100) according to claim 3, characterised in that α(p, n) represent the filter-coefficients of a first group α of filter coefficients and being defined as: 7α⁡(p,n)={ ⁢w1⁢:⁢ ⁢Δ⁡(p,n)<Th1w2⁢:⁢ ⁢Th1≤Δ⁡(p,n)<Th2p⁢:⁢ ⁢Th3≤Δ⁡(p,n)≤Th4 ⁢0⁢:⁢ ⁢otherwiseand β(p, n) represent the filter-coefficients of a second group β of filter coefficients and being defined as: 8β⁡(p,n)={ ⁢w1⁢:⁢ ⁢Δf⁡(p,n)<Th1w2⁢:⁢ ⁢Th1≤Δf⁡(p,n)<Th2p⁢:⁢ ⁢Th3≤Δf⁡(p,n)≤Th4 ⁢0⁢:⁢ ⁢otherwise wherein: Th1-Th4: are threshold values P: is a negative programmable parameter w1, w2: are positive programmable parameters; and wherein the parameters Δ and Δf are defined as follows:Δ(p,n)=|F(p+n)−F(p)|Δf(p,n)=|Ff(p+n)−F(p)|

5. The N-dimensional filter (100) according to claim 1, characterised in that the original pixel as well as the neighbour pixels p(x, y, t)T having the time component t are used either for noise reduction or motion blur removal.

6. The N-dimensional filter (100) according to claim 5, characterised in that the pixels p(x, y, t)T are either motion compensated or not.

7. The N-dimensional filter (100) according to claim 1, characterised in that the input signal is a selected spectral component of a video signal.

8. The N-dimensional filter (100) according to claim 2, characterised in that the filter coefficients α(p, n) or β(p, n) are adjusted in real time.

9. The N-dimensional filter (100) according to claim 1, characterised in that the first sign is the + sign and the second sign is the − sign.

10. The N-dimensional filter (100) according to claim 1, characterised in that the N-dimensional filter (100) is embodied in an application specific integrated circuit ASIC.

11. Method for N-dimensionally filtering an original image pixel with respect to noise reduction according to a transmission function, the method being characterised by the step of: defining at least one filter-coefficient of the transmission function such that the transmission function is capable of applying noise reduction as well as sharpness enhancement to the original image pixel input to an N-dimensional filter (100).

12. Display apparatus (200) for displaying images, comprising the N-dimensional filter (100) as claimed in claim 1.