JPH0244884A - Post filter system for removal of coding noise - Google Patents

Abstract

PURPOSE:To remove a coding noise by selecting decision threshold for a still/ dynamic area in conformity to the size of a quantization representative value. CONSTITUTION:The inputted quantization representative value Ri is inputted to a threshold table 32, and the still/dynamic area decision threshold Thi is obtained, and is turned into Thi-1 after passing through a delay circuit 33, and is inputted to 8 correction value table 38. In the correction value table 38, the value of Xi-1-X'i-1 is corrected by an inter-frame difference value Xi-1-X'i-1 coming through the delay circuit 37 and Thi-1. Then, a correction value k'(X'i-1-Xi-1) is inputted to a difference circuit 40 through the delay circuit 19 together with Xi-2. Thus, by switching the decision threshold for the still/dynamic area adaptively according to the received quantization representative value, the removal of the coding noise can be performed effectively.

Description

Detailed Description of the Invention (Technical Field of the Invention) The present invention utilizes the strong correlation that exists between adjacent pixels on a screen for image signals such as commercial television and high-definition television (high-definition). The present invention relates to a post-filter method for effectively removing coding noise generated on the receiving side in a predictive coding transmission method that performs high-efficiency coding.

(Prior Art and its Problems) First, a conventional coding method called intra-field predictive coding will be explained.

Current standard television consists of screens called frames, which are sent 30 to 25 frames per second, and each frame is made up of two consecutive fields because interlaced scanning is performed for each scanning line. It is more established. Also, the elements that make up the screen are
However, since digital processing is in mind here, one sample will be called a pixel.

Therefore, in this case, the position of each pixel within the screen depends on the sampling frequency for digitizing the signal.

FIG. 1 shows the positional relationship of each pixel 1 to 5 for this purpose. In the figure, pixel 2 is located on the left side of the current pixel 1 on the same line, pixels 3 and 4.5 are located on the upper right of pixel 1 on the line one line above in the same field,
The pixels are located directly above pixel 1 and directly above pixel 2, respectively. At this time, it is thought that there is a strong correlation between the sample values of some pixels located in the immediate vicinity, so the predicted value of the sample value XF of pixel 1? .

For example, Y I −
X 2'-------------'----
-------"--
) or ¥, =X2+3/4X4-3/4X5------
--- Create as (2), this predicted value Ma, and the true value XF
Difference e, =XF−ma, −−−−−−−−−−−
−−−−−−−−−−−−−−−−−13) is the prediction error, and by quantizing and encoding it, high-efficiency encoding is performed to reduce the number of transmission bits required. It can be done.

However, at this time, the quantized prediction error signal Q(e+)−Q (x+−father,) −X+ Yl +Δ(1r′−”−−−−−−−・
- (4) is the reproduced value X+ =Q obtained by adding quantization noise Δq1 and performing code manipulation on the receiving side.
(e+)+'Y+ -XF''-Δq + −−−−−−−−−−−−−−
−−−−−−(5) contains coding noise (quantization noise) △
q1 is superimposed.

This coding noise depends on the characteristics of the quantizer, which is determined by the number of transmission bits required by the coding.

That is, a normal image signal has 8 bits (0 to 2 bits) per pixel.
55 Leher), and the prediction error is -255 to 255
This is 9 bits. When this is encoded with 4 bits per pixel and transmitted, the number of representative quantization values is 16, 3
In the case of bits, there are eight types, and as the number of representative values decreases, quantization noise increases.

Therefore, it is especially important to remove coding noise when the required transmission rate is small.

As a method for removing this coding noise, the following post-filtering process has been performed in the prior art. That is, first, a still area and a moving area on the screen are determined as follows using the inter-frame difference absolute value f and a predetermined threshold Th.

If f, ≦Th, then stationary region f; if > Th, then moving region r + = + x + y I+ -------
---------- (6) However, Yl indicates the reproduced value of the pixel located at the same position as pixel 1 one frame before.

In the case of stationary region, X + = X I +Δq + −−−−−−−
−−−−−−−−−−−−−(7) y + = X
+''−ΔQ+' −−−−−−−−−−−−
−−−−−−−−−−−−−(8) holds, so
For example, when the filter processing shown in equation (9) is performed, the value after filtering XIF is X+F=IA (also + 1y I)
−−−−−−−−(9) The power of the difference between the true value and the filtered value is −y2E [Δq +”]−]1−−−00
) (where EC) means the average value. ), and the noise power can be reduced to half compared to the case without filtering. On the other hand, in the case of a dynamic region, Y can no longer be expressed by equation (8), and equation (9)
If the filter processing shown in is performed, the image will deteriorate.

In consideration of the above, in the conventional technology, filter processing is performed only when the area is considered to be a stationary area in the above determination. However, as mentioned above, in this method, static/moving areas are determined simply by comparing the magnitude of the inter-frame difference absolute value f and the threshold Th, and whether or not to apply a filter is determined. If it is too thick, even if it is a moving area, it may be determined to be a static area, and a filter will be applied, conversely degrading the image quality. On the other hand, if Th is set small, even a still area may be determined to be a moving area, and in this case, the filter is not applied, so the image quality improvement effect of the filter becomes small.

(Objective of the Invention) The present invention has been made in view of the above points, and adaptively switches the determination threshold for static/moving regions based on the received quantization representative value to more effectively remove coding noise. The present invention provides a post-filtering method for removing coding noise that is capable of eliminating coding noise.

The present invention will be explained in detail below.

(Principle of the Invention) In the present invention, first, a static/moving area determination threshold is determined in advance based on a quantized representative value transmitted for each pixel. For example, if the number of quantization representative values of the quantizer used is n and each quantization representative value is Ri (i=1n), the corresponding threshold Th is a function of R4 and is expressed as Th(Ri). Ru.

Compare this threshold with the absolute value of the inter-frame difference f, and if f1≦Th(R4), apply the filter F, f
If +>Th(R1), filter F2 is applied or no filter is applied.

Here, filter F means a filter in the time axis direction, and the reproduction value of one frame before -Y + (+ = 1 to m)
and the reproduction value XF of the current pixel, it is given by the following equation.

However, k is a coefficient for filter weighting, and a is a coefficient for the reproduced value of the previous frame.

For example, in the above formula, 1'4. If m=1, a, -1, F+ −+A(X+Yl) −−−−−−−−−
−1−−−−−−−−−−・−02), which coincides with the above formula (9).

On the other hand, filter F2 means a filter in the spatial direction, and is the reproduced value xi(
i=2 to n) and the current pixel reproduction value X+, it is given by the following equation.

However, h indicates a coefficient for filter weighting, and bi indicates a coefficient related to a reproduced value of a pixel in the current field.

(Validity of invention) Next, the technical background and effectiveness of the present invention will be explained.

Generally, there is a strong correlation between adjacent pixels in the same field, so the occurrence distribution of prediction error values, which is the difference between the predicted value obtained by intra-field predictive coding and the true value, is shown in Figure 2. As shown in the figure, the O concentration type occurs, and as the prediction error value increases, the frequency of occurrence decreases. Utilizing this characteristic, in quantization, the quantization width near O is made as fine as possible, and the larger the prediction error value is, the coarser the quantization width is, thereby achieving efficient reduction of coding noise. Table 1 shows 4-bit 15-level quantization characteristics as an example of this quantization.

Table With such quantization characteristics, a small amount of coding noise is superimposed on the pixels reproduced by the quantization representative value assigned to a fine quantization width, and conversely, quantization when the quantization width is coarse Large coding noise will be superimposed on the pixels reproduced using the representative value.

As a result, there is a correlation between the quantization representative value and the coding noise, and if the received quantization representative value is large, the coding noise may also be large, so even if the inter-frame difference absolute value is somewhat large. However, it can be considered that this is not caused by the dynamic region but by coding noise.

Conversely, when the quantization representative value is small, the superimposed coding noise is small, so even if the inter-frame difference is small, it can be regarded as a moving region.

From this fact, it is better to judge static/moving regions based on the absolute value of the inter-frame difference and a fixed threshold (use the information of the received quantization representative value, and set the judgment threshold when the quantization representative value is small). If the determination threshold value is set to be small, and conversely if it is large, the determination threshold value is increased, the still/moving region can be divided more effectively, and thereby the noise in the still region can be effectively removed.

Furthermore, coding noise can be removed by applying a filter in the spatial direction to a moving region instead of a filter in the time axis direction to such an extent that "blur" of the image is not visually detected.

In addition to the above considerations, the effectiveness of this postfilter is shown below from another perspective.

That is, considering the case where the edge image shown in FIG. 3 moves in the horizontal direction, the waveform shown by the solid line is taken as the edge of the current field, and the waveform shown by the broken line is taken as the edge of the field one frame before.

At this time, the area S of the shaded area is S−Σf,
−−−−−−−−−−−−−−−−−−−−−−−
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−− −−−−−−−
−−−−−−−−−−−−−−−−05) (e+ is the intra-field difference absolute value).

From this, the amount of horizontal movement ■ is It is indicated by.

From the above equation, the amount of movement ■ is not simply determined by f, but even if f is small, if e is small, there is movement, and conversely, even if f8 is large, if e is large, movement is small. Become. If this e is replaced with the quantization representative value mentioned above, when the quantization representative value is large, the judgment threshold can be increased to more accurately judge static/moving regions, and conversely, when the quantization representative value is small, the judgment threshold can be made small. It will be a good thing.

FIG. 4 shows the relationship between the quantization representative value and the inter-frame difference absolute value obtained in a simulation experiment using real image data.

In this simulation experiment, the intra-field predictive coding method using a 4-bit quantizer shown in Table 1 was used.

As shown in the figure, still images and videos [(1 pixel/frame)
and (2 pixels/frame) movement], as the quantization representative value increases, the absolute value of the inter-frame difference increases, and the determination threshold indicated by the broken line in the figure allows accurate division of static/moving regions. This enables more effective post-filter processing to remove coding noise.

(Example) A specific example of the post-filter method according to the present invention will be described below. FIG. 5 is a block diagram thereof. 11 is a difference absolute value circuit for calculating the inter-frame difference absolute value between the reproduced value of the latest pixel and the reproduced value of the pixel one frame before, 12 is a frame memory, 13 is a field line memory, and 14 is a circuit for determining the absolute value of the difference between the frames. A threshold setting circuit sets a threshold for determining a static/moving region from the quantized representative value transmitted from the transmitting side to obtain a reproduction value, and 15 is a comparison circuit that compares the absolute value of the inter-frame difference with the determination threshold. 16 is a time domain filter creation unit that removes coding noise in the time domain; 17 is a space domain filter creation unit that removes coding noise in the space domain; and 18 is a time domain filter creation unit 1.
6 and the spatial domain filter creation section 17.

FIG. 6 is an example of implementation of the post-filter method according to the present invention. Reference numeral 21 denotes a frame memory, and 22 denotes a look-up table which inputs the reproduced value and quantized representative value of the latest pixel and the reproduced value of the pixel one frame before, and can be constructed from a random access memory.

FIG. 7 shows an embodiment of this post-filter method for removing noise. The input signal of the filter is a reproduced image and its quantized representative value, and the output code is the filter output value.

In order to explain the temporal flow of signals (time chart),
The i-th reproduced image is X8, the quantization representative value is Ri, and the static/moving area determination threshold corresponding to each quantization representative value is Th, , 1
Let the reproduced image before the frame be X'i.

The input reproduced image (XF) is delayed by the two-pixel delay circuit 31, becomes x, -2, and is input to the frame memory 34. At this time, X'i 44 is read out from the frame memory 34, and the delay circuit 35 obtains the image signal X'i one frame before XF. XF and X'i are input to the difference circuit 36, and the inter-frame difference value X is obtained.

XF is obtained.

On the other hand, the input quantization representative value R1 is
2, thereby obtaining the static/moving area determination threshold Th, which passes through the delay circuit 33 and becomes Th1-, and is input into the correction value table 38.

In the correction value table 38, the value of xi-1-X (□I) is corrected as follows using the inter-frame difference values xi-1-x, which have passed through the delay circuit 37, and Th1-.

That is, when Xl-I Xl-If≦Th+-1, '(Xi
-11c'1-1). However, when O<k≦1), k=-+ k'i-+ >Tht-+, 0 is output.

The correction value '(X'i-I X 1-1) is manually inputted to the difference circuit 40 together with XF-2 via the delay circuit 19.

Signal after difference (x r -2k'(x' + -2X
i −2) ) becomes the filter output value.

The output value of this filter is X i-2k' (X i-2X i-2)-(1-to')
+(1-k)x,-2 (k=1-'), so the same temporal filter as described above can be obtained in this embodiment.

Note that in this embodiment, if the interframe difference absolute value is larger than the static/moving area determination threshold, no spatial filter is applied.

(Effects of the Invention) As described above in detail, according to the present invention, when an input image is reproduced on the receiving side of a predictive coding transmission system for an image signal that transmits a quantized representative value, changes in the input pixels When dividing moving/static regions according to the comparison result between the amount indicating the amount and the threshold value, and performing post-filter processing to remove coding noise on these divided regions, the threshold value is set to the quantization representative value. By making selections based on size, it is possible to remove coding noise extremely effectively, which has a great practical effect.

[Brief explanation of the drawing]

Fig. 1 is a pixel array pattern diagram showing an example of image signal transmission to which the present invention is applied, Fig. 2 is a characteristic diagram showing the occurrence distribution of prediction errors in intra-field predictive coding, and Fig. 3 is a detailed diagram of the present invention. A schematic diagram for explanation, FIG. 4 is a characteristic diagram showing the relationship between the quantization representative value and the absolute value of the inter-frame difference, and FIGS. 5, 6, and 7 are block diagrams showing embodiments of the present invention. . L 2.3.4.5...Pixel, 11...Difference absolute value circuit, 12.21.34...Frame memory, 13
... Field line memory, 14... Threshold value setting circuit, 15... Comparator, 16... Time domain filter creation section, 17... Spatial domain filter creation section, 18
...Switch, 22...Lookup table,
3L 33.35.3739...Delay circuit, 32.
...Threshold table, 36.40...Differential circuit, 3
8... Correction value table.

Claims (5)

[Claims] (1) Predictive coding of the image signal that transmits the quantized representative value When reproducing the input pixel on the receiving side of the transmission system, the static/moving region is divided according to the result of comparing the amount indicating the change in the input pixel with a threshold value. In the method of performing post-filter processing to remove coding noise for these divided regions, the determination threshold for static/moving regions is selected in accordance with the magnitude of the quantization representative value. A post-filter method for removing coding noise, characterized in that: (2) The post-filter method for removing coding noise according to item 1, characterized in that filters are switched based on a determination threshold value for the static/moving region. (3) The post-filter method for removing coding noise according to item 1, characterized in that a time domain filter is used for the static region. (4) The post-filter method for removing coding noise according to item 1, characterized in that a spatial domain filter is used for the motion domain. (5) Obtain the predicted value of the pixel to be encoded from the information of the pixel that has already been encoded, and obtain the quantized representative value by quantizing the difference value between the predicted value and the value of the pixel to be encoded. , the code obtained by encoding the quantized representative value is transmitted to the receiving side, and the receiving side uses the quantized representative value obtained from the code and the predicted value to determine the reproduced value of the pixel to be encoded. A pixel reproducing device on the receiving side in predictive coding transmission of a television signal that obtains a quantized representative value R of the latest pixel; When the reproduction @X@ of the latest pixel reproduced from the predicted value is obtained, the reproduction value @X@_F of the pixel one frame before the latest pixel is read from the storage unit, and the reproduction value @ X@, the quantized representative value R, and the reproduced value of the pixel of the previous frame read from the storage section@
and a post-filter section in which a filter for removing coding noise included in the reproduced value @X@ of the latest pixel is selected by X@_F. Device.