WO2008116836A2

WO2008116836A2 - Image processing method and coding device implementing said method

Info

Publication number: WO2008116836A2
Application number: PCT/EP2008/053399
Authority: WO
Inventors: Stéphane ALLIE; Xavier Ducloux; Denis Mailleux
Original assignee: Thomson Licensing
Priority date: 2007-03-23
Filing date: 2008-03-20
Publication date: 2008-10-02
Also published as: WO2008116836A3; FR2914125A1

Abstract

The invention relates to a method for processing an image divided into blocks of pixels. It comprises the following steps: - calculate (12, 30) for each one of said blocks a first value (MFk, MBFk) representative of the contrast of the contours in said block, - calculate (13, 32) for each one of said blocks a second value (MASk, MBASk) representative of the spatial activity of the said block, - calculate (14, 33) for each one of said blocks a third value (DCk, MBDCk) representative of the mean luminance of the said block, - determine (15, 34), with a view to a subsequent coding, for each one of said blocks a quantization offset step (ΔQPk) according to the said first, second and third values. The invention also relates to a coding device.

Description

IMAGE PROCESSING METHOD AND CODING DEVICE IMPLEMENTING SAID METHOD

1. Scope of the invention The invention relates to the domain of image coding. More specifically it relates to an image processing method and a coding device implementing said method.

2. Prior art Referring to figure 1 , a method for coding an image I or a sequence of images I in the form of a bitstream S, generally comprises, for each image to be coded, a step E1 to transform, with a transform such as a DCT (Discrete Cosine Transform) of luminance and chrominance data associated with the pixels of the image, a step E2 to determine at least one Target quantization step QP^* associated with the image I or with a part of this image, a step E3 of quantization to quantize the image data using one or more quantization steps and a step E4 of entropic coding of quantized data. The image quantization step QP^* is generally determined by a bitrate control algorithm as a function of a bitrate setting D determined as a function of the application and associated with the image I. According to one variant several target quantization steps QP^* are determined during step E2 for the image I, each of them associated with a group of macroblocks, such as for example a macroblock slice.

According to standard approaches, the determination step E2 of the target quantization step QP^* for an image or image slice is followed by a step E2' of local adjustment of the quantization step within said image or said slice of image, for example at the level of a block of pixels. This local adjustment is based on the recognition that the human eye perceives faults more easily on the uniform zones than on the heavily textured zones i.e. having a high spatial complexity. Consequently, the uniform zones must be preserved and therefore quantized less heavily than the textured zones that can be degraded without being perceived by the eye. From this recognition, the standard approaches modify the target quantization step QP^* locally, for example for each block of pixels from a spatial analysis of image data of these blocks with the aid of operators adapted to this effect. This spatial analysis highlights the psycho-visual characteristics of each block and enables uniform blocks and heavily textured blocks to be identified. A higher quantization step than the target quantization step QP* is then associated with the heavily textured blocks and conversely a lower quantization step than the target quantization step QP^* is associated with the uniform blocks. However, the spatial analysis operators used generally assimilate blocks that have a clear object contour, also called front, with heavily textured blocks, or assimilate blocks having a significant and continuous luminance variation with textured blocks. Consequently, a local adjustment method of the quantization step based on such operators will attribute a high quantization step both to blocks comprising contours and blocks having a continuous luminance variation, when on the contrary it should preserve these zones.

3. Summary of the invention The purpose of the invention is to compensate for at least one disadvantage of the prior art. For this purpose, the invention relates to a method of processing an image divided into blocks of pixels each of which is associated with at least one luminance value. According to the invention, the method includes the following steps: - calculate for each block a first value representative of the contrast in the block,

- calculate for each block a second value representative of the spatial activity of the block,

- calculate for each block a third value representative of the mean luminance of the block,

- determine, with a view to a subsequent coding, for each block a quantization step offset according to the first, second and third values.

The method according to the invention advantageously enables a distinction to be made between textured zones, zones including a contour and homogenous zones or zones having a distinct variation but a continuous luminance, and to attribute a quantization step offset as a function of this characterisation. According to another embodiment, wherein each block is divided into sub-blocks, the method also comprises a step to calculate for each of the sub- blocks a fourth value representative of the contrast in the sub-block, a step to calculate for each of the sub-blocks a fifth value representative of the spatial activity and a step to calculate for each of the sub-blocks a sixth value representative of the mean luminance of the sub-block. According to this embodiment, the first value is calculated, for each of the blocks, from the fourth valued of the sub-blocks of the block and of adjacent sub-blocks of the block, the second value is calculated from the fifth values of the sub-blocks of the block and of adjacent sub-blocks of the block and the third value is calculated from the sixth values of the sub-blocks of the block.

According to an advantageous characteristic, the blocks are macroblocks of 16 by 16 pixels and the sub-blocks are pixel blocks of 8 by 8 pixels. According to a particular characteristic, the first value is equal to the maximum value of the four values of the sub-blocks of the block and of the adjacent sub-blocks of the block.

According to a particular characteristic, the second value is equal to the minimum value of the fifth values of the sub-blocks of the block and of the adjacent sub-blocks of the block.

According to another particular characteristic, the third value is equal to the mean of the sixth values of the sub-blocks of the block.

According to another particular characteristic, the adjacent sub-blocks of the block have a side in common with the block.

According to a particularly advantageous characteristic, the fourth value of a sub-block of size I by J pixels is equal to

where pel,_j is the luminance value associated with the pixel of coordinate (i,j)

J-I in the sub-block (sbi) and where MAX(a ) is the maximum value Q_] of the j=0 ^* J - group {a₀, ai, ..., aj.i}. According to a particularly advantageous characteristic, the fifth value of a sub-block of size I by J pixels is equal to

- p^el>+ι,j

The invention also relates to a coding device capable of implementing the method according to the invention. The coding device comprises a transformation module to transform image data associated with each of the blocks into transform data, an image processing module suitable to determine a quantization step offset for each of the blocks, a quantization module suitable to quantize, for each of the blocks, the transform data according to a predetermined quantization step and the quantization step offset of the block and an entropic coding module suitable to code, for each block, the quantized data in the form of a bitstream. According to an essential characteristic of the invention, the image processing module comprises: - means for calculating for each one of the blocks a first value representative of the contrast in the block,

- means for calculating for each one of the blocks a second value representative of the spatial activity of the block,

- means for calculating for each one of the blocks a third value representative of the mean luminance of the block, and

- means for determining for each one of said blocks said quantization step offset (ΔQP_k) according to said first, second and third values.

4. List of figures The invention will be better understood and illustrated by means of embodiments and implementations, by no means limiting, with reference to the figures attached in the appendix, wherein:

- figure 1 illustrates an image coding method according to the prior art,

- figure 2 illustrates an image processing method according to a first embodiment of the invention,

- figure 3 illustrates a step of the image processing method according to the first embodiment of the invention, - figure 4 illustrates an image processing method according to a second embodiment of the invention,

- figure 5 illustrates a step of the image processing method according to the second embodiment of the invention, - figure 6 represents a macroblock and the adjacent blocks of said macroblock, and

- figure 7 illustrates an image coding device according to the invention, and

- figure 8 illustrates an image sequence coding device according to the invention.

5. Detailed description of the invention

The invention relates to an image processing method to locally adjust a target quantization step QP^* within an image or an image slice made of pixels, each of which is associated with at least one luminance value. A first embodiment is described with reference to figure 2. According to this embodiment, the image divided into N blocks bk not overlapping with k the index of the block in the image varying from 0 to N-1. The method includes an initialisation step 10 at the value 0 of the index k of the blocks bk of said image. It then comprises a calculation step 12 for the block bk of a first value noted as MF_k representative of the contrast in the block bk. More precisely, this first value is representative of the maximum contrast of block bk in the vertical direction and the horizontal direction. The value MF_k is preferentially calculated according to the following equation:

where: - 1 is the height of block bk in number of pixels,

- J is the width of block bk in number of pixels,

- peli_j is the luminance value associated with the pixel of coordinate (i,j) in the block bk,

J-I

- MAX(a_}) is the maximum value Q_] in the group {ao, ai, ..., aj.-i}.

Generally, I =J=8. It also includes a calculation step 13 for the pixel block bk of a second value noted as MASk representative of the spatial activity in the block bk. The value of MASk is preferentially calculated according to the following equation:

It also includes a calculation step 14 for the pixel block bk of a third value noted as DCk representative of the mean luminance also known as average level of luminance in the block bk. The value of DCk is preferentially calculated according to the following equation:

Steps 12, 13 and 14 of the method can also be performed simultaneously or in an entirely different order to the order described.

The combination of the three values determined in steps 12 to 14 advantageously enables the contents of block bk to be characterized. They particularly make it possible to determine if the block bk includes a contour, if it is textured, if it is homogenous or if it includes a zone of significant variation but with continuous luminance. In fact, a homogenous block bk, a block bk comprising a zone of significant variation but continuous luminance or including a clear and isolated contour should be preserved, i.e. quantized with a quantization step lower than the target quantization step QP*, whereas a textured block bk can be quantized with a higher quantization step. Further, the DCk criteria enables the luminance level of the homogenous zones to be characterized. Indeed, in a homogenous zone, the faults are visible except when the mean luminance level of the zone is high or conversely low.

At step 15, the target quantization step QP^* associated with the image or image slice to which the block bk belongs is adjusted at the level of block bk as a function of the values MF_k, MASk and DCk calculated during steps 10 to 14. More specifically, during step 15 a quantization offset value noted as ΔQP_k, is determined with respect to QP^* for the block bk, i.e. the quantization step later used to quantize the block b_k is equal to QP^*+ ΔQP_k. According to an embodiment described with reference to figure 3, the quantization offset ΔQP_k is determined in step 15 in the following manner: At step 150, MFk is compared to a first threshold S1 , if (MF_k > S1 ) at step 151 , MAS_k is compared to a second threshold S2, if (MASk > S2) then at step 152, ΔQP_k = ΔQP1 , with ΔQP1 a predetermined positive integer, else, at step 153, ΔQP_k = ΔQP2, with ΔQP2 a predetermined negative integer, else at step 154, MASk is compared to a third threshold S3 and DCk is compared to a fourth threshold S4 and a fifth threshold S5 if (MASk < S3 AND DC_k > S4 AND DC_k < S5) then at step 155, ΔQP_k = ΔQP3, with ΔQP3 a predetermined negative integer, else at step 156, MASk is compared to a sixth threshold S6, if (MASk > S6) then at step 157, ΔQP_k = ΔQP1 , else at step 158, ΔQP_k = 0.

In fact, at step 152 and step 157, the content of block bk is characterised as having a textured content that can be further quantized. At step 153, the content of block bk is identified as containing a marked contour that must be maintained. At step 155, the content of block bk is characterised as having homogenous content (i.e. uniform or low contrast) having an average luminance level, i.e. between S4 and S5, that must be maintained. At step 158, the content of block bk is not characterised as being textured, or uniform with an average luminance level or as containing an isolated colour, which is why the value 0 is attributed. Preferentially, ΔQ3 is chosen equal to ΔQ2. For example, ΔQ3=ΔQ2=1.

At step 16, the k index is incremented by 1. At step 18, k is compared to the value N. If k is greater than or equal to N then the method is terminated, otherwise the method restarts at step 12.

According to the method of the invention, the determination of the value of the quantization offset ΔQP_k does not require knowledge of the value of the target quantization step QP*, which is generally determined during coding by a bitrate control method. Consequently, the quantization offset values ΔQP_k can be determined prior to the coding method.

The method described for one image can be applied to a sequence of many images, e.g. a video. However, most video coding standards authorise modification of the quantization step only at a macroblock level. A macroblock is a pixel block. In general, a macroblock is a block of pixels of 16 by 16. It is itself divided into several non-overlapping sub-blocks of 8 by 8 pixels. For this purpose, another embodiment is described with reference to figure 4. The image is divided into N non-overlapping macroblocks, noted as MB_k, where k is the macroblock index in the image, k varying from 0 to N. Each sub-block is noted as sbi, where I is the sub-block index in the image, I varying from 0 to M. The method according to the second embodiment comprises an initialisation step 20 with a value 0 of the k index of the macroblocks MB_k and of the I index of the sub-blocks sbi of said image. It then comprises a calculation step 22 for the sub-block of pixels sbi of a first value noted as MF_k representative of the contrast in the sub-block sbi. More precisely, it is representative of the maximum contrast of the sub-block b_k in the vertical direction and the horizontal direction. The value MFi is preferentially calculated according to the following equation:

where: - 1 is the height of the sub-block sbi in number of pixels,

- J is the width of the sub-block sbi in number of pixels,

- peli_j is the luminance value associated with the pixel of coordinate (i,j) in the sub-block sbi,

J-I

- MAX(U₁) is the maximum value a, in the group {a₀, ai, ..., aj.-i}.

It also includes a calculation step 24 for the sub-block of pixels sbi of a second value noted as MASi representative of the spatial activity in the sub- block sbi. The value MASi is preferentially calculated according to the following equation:

^MASι ^~P^el _1+lj

It also includes a calculation step 25 for the block of pixels sbi of a third value noted as DCi representative of the mean luminance in the block sbi. The value DCi is preferentially calculated according to the following equation:

Steps 22, 24 and 25 of the method can also be performed simultaneously or in an entirely different order to that described. At step 26, the I index is incremented by 1.

At step 28, I is compared to the value M. If I is greater than or equal to M then the method continues to step 30, otherwise the method restarts at step 22.

At step 30, the MFi values calculated for each sub-block sbi of a macroblock MB_k are combined to obtain an MBF_k value for the macroblock MBk. The value MBF_k is the maximum value MFi among the values MFi associated with the sub-blocks sbi of the macroblock as well as the adjacent sub-blocks sbi of the macroblock MB_k, i.e. those that have a commons side with one of the sub-blocks sbi of the macroblock MB_k. The sub-blocks sbi adjacent to macroblock MB_k are greyed out on figure 4. This selection advantageously enables priority to be given to the neighbouring area of a contour. At step 32, the MASi values calculated for each sub-block sbi of a macroblock MB_k are combined to obtain an MBAS_k value for the macroblock MB_k. Le MBASki value is the minimum MASi value among the MASi values associated with the sub-blocks sbi of the macroblock MB_k as well as with the sub-blocks sbi adjacent to the macroblock MB_k, i.e. the greyed out sub-blocks sbi in figure 4. This selection advantageously enables the zone to favour to be enlarged beyond the uniform zone and a particularity of spatial activity to be suppressed, for example.

These two filtering operations adjacent to the macroblock MB_k allow the quantization step offset values in the image to be smoothed i.e. from one macroblock to the next, to avoid sudden breaks in quantization that could create visible effects. Other filters can be applied. Notably it is possible to imagine stretching the neighbourhood to non-adjacent sub-blocks. The DCi values calculated for each sub-block sbi of a macroblock MB_k are also combined to obtain a mean value MBDCk for the macroblock MB_k. For example, MBDCk is the mean of the DCk values associated with each sbi sub-block of the macroblock MB_k. At step 34, the target quantization step QP* associated with the image or image slice to which belongs the macroblock MB_k is adjusted according to the values MBF_k, MBAS_k and MBDC_k calculated during steps 30 to 32. For this purpose, an offset quantization value with respect to a QP*, noted as ΔQP_k, is determined for the macroblock MB_k. The quantization step later used to quantize the macroblock MB_k. is equal to QP*+ ΔQPk. According to an embodiment described with reference to figure 6, the quantization offset is determined in step 34 in the following manner: at step 350, MBF_k is compared to a first threshold SMB1 , if (MBF_k > SMB1 ) at step 351 , MBASk is compared to a second threshold SMB2, if (MBASk > SMB2) then at step 352, ΔQP_k = ΔQP1 , with ΔQP1 a positive integer, else, at step 353, ΔQP_k = ΔQP2, with ΔQP2 a negative integer, else at step 354, MBAS_k is compared to a third threshold SMB3 and MBDCk is compared to a fourth threshold SMB4 and a fifth threshold SMB5 if (MBASk < SMB3 AND MBDC_k > SMB4 AND MBDC_k< SMB5) then at step 355, ΔQP_k = ΔQP3, with ΔQP3 a negative integer, else at step 356, MBAS_k is compared to a sixth threshold SMB6, if (MBASk > SMB6) then at step 357, ΔQP_k = ΔQP1 , else at step 358, ΔQP_k = 0.

In fact, at step 352 and step 357, the content of macroblock MB_k is characterised as having a textured content that can be further quantized. At step 353, the content of block MB_k is identified as containing a marked contour that must be preserved. At step 355, the content of macroblock MB_k is characterised as being a uniform content having an average luminance level, i.e. between S4 and S5, that must be preserved. At step 358, the contents of macroblock MB_k are not characterised as being textured, or uniform with an average luminance level or as containing an isolated colour, which is why the value 0 is attributed.

Preferentially, ΔQ3 is chosen equal to ΔQ2 for example equal to 1. At step 36, the kl index is incremented by 1.

At step 38, k is compared to the value N. If k is superior to N then the method is terminated, otherwise the method restarts at step 30. According to the method of the invention, the determination of the value of the quantization offset ΔQP_k does not require knowledge of the value of the target quantization step QP*, which is generally determined during coding by a bitrate control method. Consequently, the quantization offset values ΔQP_k can be determined prior to the coding method. This embodiment described for a macroblock MBk divided into sub- blocks sbi can be applied more generally to all blocks of pixels b_k divided into sub-blocks sbi. For example, this embodiment can be applied to blocks b_k of size 8 by 8 pixels divided into sub-blocks sbi of size 4 by 4 pixels.

In reference to figure 7, the invention relates to an coding device ENC of an image I in the form of a bitstream S suitable to implement the method of the invention. It comprises a transformation module T able to receive the image I. It further comprises an image processing module P able to receive the input image I at the input, a quantization module Q whose first input is linked to the output of the processing module T and an entropic coding module C whose input is linked to the output of the quantization module Q. The entropic coding module C is able to generate the bitstream S. The transformation module T is suitable to transform image data e.g. chrominance and luminance data, associated with the pixels of the image I, into transform data. This module implements for example, a DCT type transform. The image processing module P is adapted to implement the steps 10 to 18 and/or 20 to 38 of the method according to the invention. In particular, it is adapted to determine for each block b_k or macroblock MB_k within the image I, a quantization offset ΔQP_k with respect to a target quantization step QP* associated with the image I or an image slice to which belongs the block bk, respectively the macroblock MB_k. The module Q is adapted to quantize the transform data into quantized data using a quantization step possibly locally adjusted, i.e. equal to QP^*+ ΔQP_k. The coding module C is adapted to implement an entropic coding of quantized data in order to generate the bit stream S at the output of the ENC device.

In reference to figure 8, the invention relates to a coding device ENCV of a sequence of images V in the form of a bitstream SV suitable to implement the method of the invention. The coding device ENCV comprises modules already described in reference to figure 6 that have the same references i.e. the modules T and Q. It also comprises a subtractor D able to receive on a first input images I of the sequence V and on a second input images or part of a prediction image Ic. The input of the transformation module T is linked to the output of the subtractor D. It also includes, an image processing module P able to receive at the input images I from the sequence V. The output of the image processing module P is linked to the input of the quantization module Q. In particular, it is suitable to determine for each block bk or macroblock MB_k within an image I, a quantization offset ΔQP_k with respect to a target quantization step QP^* associated with an image I or with an image slice to which the block bk or the macroblock MB_k respectively belongs. The target quantization step QP^* is generally determined by a bitrate control module, not shown on the figure 8. The image or part of the prediction image IC is generally obtained by motion compensation of a stored image in the MEM memory. For this purpose, the ENCV coding device includes a motion compensation module MC with an input linked to the output of a motion estimation module ME. The image of part of the predicted image IC can possibly be obtained by spatial prediction of an image stored in the memory MEM. For this purpose, the coding device ENCV includes a spatial prediction module INTRA. The selection of the prediction mode is performed by the selection module SW. The coding module ENCV also includes an inverse quantization module IQ whose input is linked to the output of the quantization device Q, an inverse transformation module IT whose the input is linked to the output of the inverse quantization device IQ. The ENCV coding device includes a adder A with two inputs, one being linked to the output of the inverse transformation module IT and the other being linked to the output selection module SW. The output of the adder A is linked to a memory MEM suitable to store images at the output of the adder A. The motion estimation module ME is suitable to estimate the motion vectors MV between a current image I and an image previously coded. The input of the motion compensation module MC and possibly the input of the spatial prediction module INTRA are also linked to the output of the memory MEM. Finally the device ENC comprises an entropic coding module CV whose first input is linked to the output of the quantization module Q and whose second input is linked to the output of the motion estimation module ME. The entropic coding module CV is suitable to code the image data at the output of the quantization module Q and the motion vectors MV in the form of the bitstream SV. The modules ME, MC, INTRA, IT, IQ and C are well known by those skilled in the art and are not further described. These modules are described in the video coding standards MPEG-2 or H.264.

According to a variant, the image processing module P is external to the coding device. It is for example integrated into a pre-processing device itself adapted to process images or a sequence of images prior to the coding with a view to facilitating the processing. The image processing module P comprises a means of storing said quantization offset values ΔQP_k. The quantization offset values ΔQP_k can for example be stored in a Look Up Table form associating a quantization offset value ΔQP_k with each block b_k or macroblock MB_k.

According to an advantageous embodiment the coding device ENCV codes a sequence of images in accordance with the video coding standard H264 known as MPEG-4 part 10 or MPEG-4 AVC.

Claims

1. Method for processing an image divided into blocks of pixels each of which is associated with at least one luminance value characterized in that it comprises the following steps:

- calculate (12, 30) for each one of said blocks a first value (MF_k, MBF_k) representative of the contrast in the said block, said value being calculated on the basis of values representative of the contrast of blocks adjacent to said block;

- calculate (13, 32) for each one of said blocks a second value (MAS_k, MBAS_k) representative of the spatial activity of the said block, said value being calculated on the basis of values representative of the contrast of blocks adjacent to said block;

- calculate (14, 33) for each one of said blocks a third value (DC_k, MBDC_k) representative of the mean luminance of the said block,

- determine (15, 34), with a view to a subsequent coding, for each one of said blocks a quantization offset step (ΔQP_k) according to the said first, second and third values.

2. Method according to claim 1 , wherein each one of said blocks (b_k, MB_k) being divided into sub-blocks (sbi), said method comprises a step to calculate (22) for each of the sub-blocks a fourth value (MFi) representative of the contrast in said sub-block (sbi), a step to calculate (24) for each of the sub- blocks a fifth value (MASi) representative of the spatial activity of said sub- block (sbi) and a step to calculate (25) for each of the sub-blocks a sixth value (DCi) representative of the mean luminance of said sub-block (sbi) and wherein said first value (MBF_k) is calculated, for each of said blocks (b_k, MB_k), from said fourth values of said sub-blocks (sbi) of said block (b_k, MB_k) and of adjacent sub-blocks (sbi) of said block (b_k, MB_k), wherein the said second value (MBAS_k) is calculated from said fifth values of the sub-blocks (sbi) of said block (b_k, MB_k) and of adjacent sub-blocks (sbi) of said block (b_k, MB_k), and wherein said third value (MBDCk) is calculated from said sixth values of said sub-blocks (sbi) of the said block (bk, MB_k).

3. Method according to claim 2, wherein said blocks (MB_k, bk) are macroblocks of size 16 by 16 pixels and said sub-blocks (sbi) are blocks of pixels of size 8 by 8 pixels.

4. Method according to claim 2 or 3, wherein said first value (MBF_k) is equal to the maximum value of said fourth values (MFi) of the sub-blocks (sbi) of said block (MBk, bk) and of the adjacent sub-blocks (sbi) of said block (MB_k, bk).

5. Method according to one of claims 2 to 4, wherein said second value (MBASk) is equal to the minimum value of said fifth values (MASi) of the sub- blocks (sbi) of said block (MB_k, b_k) and of the adjacent sub-blocks (sbi) to said block (MBk, b_k).

6. Method according to one of claims 2 to 5, in which said third value (MBDCk) is equal to the mean of said sixth values (DCi) of sub-blocks (sbi) of said block (MBk, b_k).

7. Method according to one of claims 2 to 6 in which said adjacent sub-blocks (sbi) of said block (MB_k, b_k) are the sub-blocks (sbi) having a side in common with said block (MB_k, bk).

8. Method according to one of claims 2 to 7, in which said fourth value (MFi) of a sub-block (sbi) of I by J pixels is equal to

J-I in the sub-block (sbi) and where MAX(a ) is the maximum value a_j of the j=0 ^J group {a₀, ai, ..., aj.i}.

9. Method according to one of claims 2 to 8, in which said fifth value (MASi) of a sub-block (sbi) of I by J pixels is equal to

- p^el>+ι,j

where pel,_j is the luminance value associated with the pixel of coordinate (i,j) in the sub-block sbi.

10. Coding device of an image (ENC) or a sequence of images (ENCV), each image being divided into blocks of pixels (bk, MB_k) each of which is associated with at least one luminance value, said coding device (ENC, ENCV) comprising a transformation module (T) to transform image data of each one of said blocks (bk, MB_k) into transform data, an image processing module (P) suitable to determine for each of said blocks a quantization step offset (ΔQP_k), a quantization module (Q) suitable to quantize for each of said blocks, said transform data according to a predetermined quantization step (QP^*) and the quantization step offset (ΔQP_k) of said block, and an entropic coding module (C, CV) suitable to code, for each of said blocks, said quantized data in the form of a bit stream (S SV), said coding device (ENC, ENCV) being characterized in that said image processing module (P) comprises:

- means for calculating for each one of said blocks a first value (MFk, MBF_k) representative of contrast in the said block, said value being calculated on the basis of values representative of the contrast of blocks adjacent to said block,

- means for calculating for each one of said blocks a second value (MASk, MBASk) representative of the spatial activity of the block said value being calculated on the basis of values representative of the spatial activity of blocks adjacent to said block,

- means for calculating for each one of said blocks a third value (DCk, MBDCk) representative of the mean luminance of the block, and