Method of detecting blocking artefacts

Method of detecting block artifacts

The present invention relates to a method for processing data corresponding to pixels of a digital image sequence in order to detect grids corresponding to blocking artefacts, the method comprising a step of high-pass filtering a part of the digital image to provide at least one A card of discontinuous pixels, and a step of detecting block artifacts based on the card of at least one discontinuous pixel.

The invention also relates to a television receiver including a processing device that executes the data processing method according to the invention.

Especially when detecting block artifacts in digital images and correcting the data contained in these blocks, the present invention can be applied to reduce visual artifacts caused by block-based coding techniques, which have been based on block-based coding techniques, such as MPEG The standard ("Moving Picture Experts Group") first encodes and then decodes the digital image.

Block artifacts constitute a crucial issue for block-based coding techniques that use discrete cosine transform DCT-type discrete transforms. They appear in the form of block mosaics, and block mosaics are sometimes particularly conspicuous in the decoded image sequence. These artifacts are caused by the strong quantization after the discrete transformation, which makes strong discontinuities appear on the boundaries of the coded blocks.

International patent application WO01/20912 (Case No: PHF99579) describes a method capable of detecting and locating grids corresponding to block artifacts in decoded digital images. This method authorizes the detection of three regular grid sizes, namely 8×8, 10×8, and 12×8 pixels. The grid size is determined by the main format of the image used for broadcast television digital programs. The grid 8×8 corresponds to the image sequence coded in the format of 720 pixels with 576 lines, and the grid 10-11-11×8 (approximately 10×8 grid) corresponds to the coding in the 576×540 format (called It is the coding format 3/4) and the grid 12×8 corresponds to the coding of the 576×480 format (referred to as the coding format 2/3). The size of the grid is obtained by searching for the most common distance between block artifacts. The offset of the grid relative to the image origin (0, 0) is obtained by searching for the existence of the largest number of block artifacts among all possible offsets.

An object of the present invention is to provide a more effective data processing method.

The prior art method is based on searching and detecting block artifacts regularly spaced apart. Therefore, it only searches for a grid size and a grid offset relative to the origin of the image within the image. However, due to the resampling of the image, the grid may be distorted within the image. Sometimes this distortion may be known in advance, such as in the case of the 3/4 encoding format, where the width of the grid varies with the 10-11-11 mode. However, this change is usually arbitrary because it originates from, for example, rate transcoding, image format conversion in 16/9 television receivers (for example, from 4/3 format to 16/9 format), image parts Scaling, A/D conversion, or even a combination of these different conversions. In this case, the prior art method maintains the grid and position of the most common size, and applies a post-processing step based on this grid at the risk of partial or even inefficient correction.

In order to cope with these problems, the data processing method according to the present invention is characterized in that the method includes searching within a portion of the image for a group of grid lines having a block artifact density substantially greater than that of its adjacent lines .

The processing method according to the present invention is based on an analysis and partitioning of the block artifacts of each line of the image, rather than the periodic search based on the distance between the block artifacts as in the prior art. The result is a set of grid rows, rather than a grid with fixed-size meshes, in which the distance between the grid rows can be changed due to the resampling of the image. Therefore, the processing method according to the present invention provides the possibility of processing the re-sampled image without first knowing the possible re-sampling operation.

Moreover, compared with the method of selecting one segment by segment in the prior art, since the block artifacts are selected line by line, the risk of false detection is reduced and the efficiency of the method is significantly improved.

With reference to the embodiments described below, these and other aspects of the present invention will be made apparent and illustrated by non-limiting examples.

In the drawings: Fig. 1 shows the data processing method according to the present invention; Fig. 2 illustrates two illusion distribution graphs p1 and p2 expressed in the spatial domain and the frequency domain. In principle, these illusion distribution graphs appear on the basis of In an image coded by block-based coding technology; Figure 3 describes the method of correcting block artifacts; and Figure 4 describes the principle of correcting p2 type block artifacts.

The present invention relates to a method of processing a sequence of digital images encoded and decoded according to a block-based encoding technique. In this example, the encoding technique used is the MPEG standard based on discrete cosine transform DCT, but alternatively, any other equivalent standard may be applied, such as, for example, the H.263 or H.26L standard. It should be noted that this method can also be applied to, for example, static images (fixed images) encoded according to the JPEG standard. The processing method first involves detecting block artifacts caused by these block-based coding techniques, and then involves subsequent applications such as, for example, post-processing techniques or image quality measurement.

Fig. 1 graphically shows the processing method according to the present invention. This method first includes the step of high-pass filtering a part of the HPF (110) digital image. If the image is composed of two interlaced frames, the part is, for example, one of the two fields of a frame. In a preferred embodiment, the high-pass filtering step is a gradient filtering step using the filter hp1=[1, -1, -4, 8, -4, -1, 1]. Apply the filter horizontally and vertically one line by line LGN to the pixels of the sequence of digital image field FLD whose brightness is Y(m, n), where m and n are integers between 1 and M and 1 and N, respectively, And respectively correspond to the pixel position in the domain according to the vertical and horizontal axes (for example, in the 576×720 encoding format, M=288 and N=720).

The result of the filtering operation is preferably composed of two images of discontinuous pixels, that is, a horizontal image Eh and a vertical image Ev. If the resampling operation is mainly performed in the horizontal direction, the horizontal map Eh showing the vertical discontinuity satisfies the first approximation. However, when the processing method according to the present invention is based on processing two images Eh and Ev of discontinuous pixels, the method will have the best efficiency.

Other gradient filtering may be, for example, the high-pass filter for wavelet transformation hp2=[0.045635882765054703, -0.028771763667464256, -0.2956358790397644, proposed by Antonini et al. in the article "Image Coding Using Wavelet Transform" 0.5574351615905762, -0.2956358790397644, -0.028771763667464256, 0.045635882765054703], IEEE Trans. Image Processing, Volume 1, No. 2, pages 205-220, April 1992. Moreover, it is possible to implement the high-pass filter hp1 in a particularly simple way and produce a result close to that of the filter hp2.

The method includes a step (120) of determining the discontinuity corresponding to the block artifact BAD. In fact, discontinuities can correspond to block artifacts as well as natural contours. The selection of pixels corresponding to the block artifact is performed as a function of the value of the filter coefficient Yf corresponding to the discontinuous pixels, resulting in two binary maps of the approximate location of the basic block artifact. Figure 2 illustrates the two artifact distribution maps p1 and p2 in the spatial domain and the frequency domain after filtering with the filter hp1 or hp2. In principle, these two artifact distribution maps appear in the image coded according to the block coding technology. in. The first distribution map p1 corresponds to a standard block artifact, and the second distribution map p2 corresponds to a block artifact that appears in an image that has undergone a resampling operation or an equivalent processing operation. In the spatial domain, the first profile p1 is a single step, and the second profile p2 is a double step. In the frequency domain, the first profile p1 is represented as a peak, and the second profile p2 is represented as a double peak.

In a preferred embodiment, the step of determining the discontinuity corresponding to the block artifact includes a sub-step of detecting natural contours and invisible artifacts. For this reason, the coefficient values of the horizontal Yfh(m,n) and/or vertical Yfv(m,n) filtering must be located between the two thresholds in order to be able to correspond to the block artifacts. The first threshold S1 corresponds to the visibility threshold, and the second threshold corresponds to the limit, according to which the pixel position (m, n) corresponds to a natural contour. Preferably, the following conditions are used for the absolute value of the filter coefficient: S1<|Yfh(m,n)|<S2 and S1<|Yfv(m,n)|<S2 as replacements, using the following conditions: S1<|Yfh(m ,N)|2+|Yfv(m,n)|2<S'2 where S'1 and S'2 have the same functions as S1 and S2. The threshold value depends on the filter used. For the filter hp2, for example, S'1=0.6 and S'2=400, S1=0.5 and S2=20 are used. It is particularly advantageous that in the case of applying MPEG-4, it is possible to access the video data stream and therefore the domain quantization step, and the threshold S1 is changed to the threshold S2 as a function of the quantization step in order to further improve the efficiency of the processing method . For example, the threshold value is a linear function of the quantization order.

The step of determining the discontinuity corresponding to the blocking artifact further includes the sub-step of detecting the blocking artifact. If the following conditions are met, scan the vertical map Ev in the horizontal direction corresponding to row m to detect the vertical artifacts corresponding to the distribution map p1: |Yfv(m,n)|>|Yfv(m,n+k)|k =-2, -1, +1, +2.

If |Y(m,n)-Y(m,n-1)|<|Y(m,n)-Y(m,n+1)|, the boundary of the block is located at the pixel position (m,n) and Between the pixel position (m, n+1), on the contrary, the boundary is located between the pixel position (m, n-1) and the pixel position (m, n).

If the following multiple conditions are met, detect the artifact corresponding to the distribution map p2: fl·|Yfv(m,n)|<(|Yfv(m,n-1)|+|Yfv(m,n+1)|) |Yfv(m,n-1)|>f2·|Yfv(m,n-2)||Yfv(m,n+1)|>f2·|Yfv(m,n+2)|

Wherein in the preferred embodiment f1=6 and f2=2.

The boundary of the block is located between the pixel position (m, n-1) and the pixel position (m, n). In the same way, the horizontal artifacts corresponding to each of the profiles p1 and p2 are detected by scanning the horizontal map Eh including the coefficient Yfh(m, n) filtered in the vertical direction corresponding to the column n.

In another easy-to-implement embodiment, the high-pass filtering step is based on a gradient filtering operation using a filter hp3=[-1,1]. This type of filter provides the possibility to easily detect the standard type of block artifacts corresponding to the profile p1. The step of determining the discontinuity corresponding to the block artifact includes the sub-step of detecting natural contours, such as natural contours detected when the following conditions are met: |Yfh(m,n)|<Sh, |Yfv(m,n)|< Sv where in the example, for the brightness value Y(m, n) varying between 0 and 255, Sh=35 and Sv=50.

The step of determining the discontinuity corresponding to the block artifact includes the sub-steps of detecting the block artifact on the Yfh and/or Yfv filter values of pixels other than the natural contour, and detecting the block artifact when the following conditions are met : |Yfh[i,j]|>|Yfh[i,j-1]|+|Yfh~|2|Yfh[i,j]|>|Yfh[i,j+1]|+|Yfh~| 2]]>| Yfh| is the average value of the absolute value of Yfh in the domain.

The processing method also includes a step of searching (130) pixel rows having a higher density of basic block artifact fragments in the current domain than adjacent rows.

The search step first includes a selection sub-step SEL (131) to select a segment in the horizontal or vertical line of the discontinuous pixel map, the segment including a large number of continuous block artifacts higher than a predetermined threshold S0. In fact, isolated discontinuities usually correspond to complementary noise, while block artifacts produced by coarse quantization of DCT coefficients usually produce linear errors along the coded block. The predetermined threshold value S0 does not need to be too low, so as to be unfavorable for error detection. But it does not have to be too high, lest it unduly limit the selection by reducing the number of pieces of basic block artifacts detected. In fact, the value S0 is fixed at 3 for a field of 720 pixels in 288 rows.

The search step also includes the sub-step of calculating the level Nbi of each line of Li block artifacts in CAL (132), where i is an integer corresponding to the number of lines in the domain. In a preferred embodiment, the level of block artifacts is obtained by counting the number of pixels associated with the basic artifact fragments present in the row. Alternatively, the level of block artifacts can be obtained by summing the values of the filter coefficients Yf of discontinuous pixels, which correspond to the basic artifacts of the selected segments in the row.

The search step finally includes the sub-step of determining GLD (133) grid rows, where each grid row is detected by comparing with a set of adjacent rows.

In the case of the first profile p1, by comparing the block artifact levels of the current row Li with the immediately preceding row Li-1 and the immediately following row Li+1, if the following conditions are met, the row Li Determined as a line of the grid: Nbi>α(Nbi-1+Nbi+Nbi+1), Nbi>T1·N where α is, for example, a coefficient equal to 2/3 when used for detecting vertical lines, and when used for detecting When the horizontal line is a factor equal to 3/5; T1 is the minimum percentage of artifacts in a line, and this percentage can be used to consider the line attributed to the grid. In this example, the percentage used is equal to 10%, where N is each line The number of pixels is 720 pixels in this example, so that the product T1.N is equal to 72.

In the case of the second profile p2, by comparing the block artifact levels of the current row Li with the immediately preceding rows Li-1 and Li-2, and the immediately following rows Li+1 and Li+2, if it is satisfied The following conditions determine the row Li as a row of the grid: Nbi>β(Nbi-2+Nbi-1+Nbi+Nbi+1+Nbi+2), Nbi>T2·N where β is in this example A coefficient equal to 2/3; T2 is the minimum percentage of artifacts in a row, which is equal to 5% when used to detect vertical lines and 20% when used to detect horizontal lines. The condition Nbi>T2·N provides the possibility of controlling the reliability of the system. By increasing the value of T2, the risk of false detection can be reduced.

The steps of the above processing method can be applied to the set of rows of the domain (150). Then, the processing method includes the step of verifying the GV (140) to determine whether there are a large number of grid rows in the domain. Once the entire domain has been investigated, this step will occur. This verification step ensures that there will be no erroneous detection of the rows of the grid, especially for image sequences that have not been coded according to block-based coding techniques and then decoded or have been coded according to these techniques at a high bit rate. In fact, if the number of grid rows is found to be very small in the same domain, false detection may occur. Therefore, the verification step includes comparing the total number of grid rows found in the domain Ntot, which is equal to the sum of the number of horizontal grids and the number of vertical grids, and the sum has a predetermined threshold Stot. In this example, the number Ntot only totals the grid rows corresponding to the distribution map p1; however, it is also possible to consider the grid rows corresponding to the distribution map p2 by modifying the value of the threshold Stot. If the total number Ntot is higher than the threshold Stot, there is a grid in the domain. By assuming that the maximum size of the grid is 16×16 pixels, the predetermined threshold Stot is a function of the horizontal and vertical dimensions H and V of the domain, and in order for the detection to be effective, at least a small part of the grid must be detected, in this example Use a value equal to one-third, in other words: Stot=(H+V)/(3×16) The first application of the data processing method according to the present invention consists of post-processing images to correct in the rows of the grid The illusion of existence. The correction depends on the profile of the detected block artifacts. If the block artifact corresponds to the profile p1, the correction described with reference to Fig. 3 is applied. The method for correcting block artifacts includes the following steps:-calculating the first discrete cosine transform DCT1 (31) of the first set of N data u, the set of data is located on the left or above the block boundary;-calculating the second set of N data v The second cosine discrete transform DCT1 (32), the group of data is located on the right or below the block boundary and the group is adjacent to the first group;-calculate a group corresponding to the first and second groups of concatenated CON (30) The total discrete cosine transform DCT2 (33) of 2N data w and provide a set of transformed data W;-Determine PRED (34) according to the transformed data U and V obtained by the first (31) and second (32) transform DCT1 The predicted maximum frequency kwpred is calculated in the following way: kwpred=2.max(kumax, kvmax)+2 and kumax=max(k{0,...

-Set the odd transformed data W to zero according to the total discrete transformation, and generate correction data W'to correct ZER (35), the frequency of the transformed data is higher than the predicted maximum frequency;-Calculate the inverse discrete cosine transform of the corrected data IDCT2 (36), which generates the filtered data W'intended to be displayed on the screen next.

If the block artifact corresponds to the profile p2, the correction must be modified significantly. In fact, as shown in the example of Fig. 4, since the double step corresponding to the distribution map p2, the position of the block boundary must be given more accurately. To this end, the correction method includes the step of adjusting the brightness value of the intermediate pixel p(n) in order to provide the brightness value of the p(n+1) pixel directly to the right of the brightness value. The previously described steps are then applied, where the boundary of the block appears to the left of the middle pixel, which then forms part of the segment v. Alternatively, it is possible to select the brightness value of the middle pixel to correspond to the brightness value of the left pixel, or to correspond to the brightness value of the pixel having the closest brightness value. In both cases, the positioning of segments u and v are adjusted accordingly in order to apply the correction step.

The second application of the data processing method according to the present invention is constituted by a device for measuring the block level of the domain in order to determine the quality of the image. The result of this measurement is also called a metric measurement, and the result is calculated by adding the level Nbi of the block artifacts of different rows of the grid, for example, domain by domain.

It is possible to implement the processing method according to the invention by means of a television receiver circuit, which circuit is adapted to be programmed. The computer program stored in the programming memory allows the circuit to perform different operations as described above with reference to FIG. 1. It is also possible to load a computer program into the programming memory for reading a data carrier such as, for example, a disk containing the program. The reading operation can also be performed through a communication network such as, for example, the Internet. In this case, the service provider will hand over the downloadable signal form of the computer program to those interested in it for free processing.

Any reference signs between parentheses in this document should not be restrictive. The use of the verb "comprise" and its conjugations does not exclude the existence of elements or steps other than those stated in the claims. The use of the article "a" or "an" before an element or step does not exclude the presence of multiple such elements or steps.