KR100203713B1 - Method and apparatus for processing an image signal having an object by using an extension-interpolation technique - Google Patents

Abstract

The present invention is a method for converting a processing block included in an image frame signal including an object by using a tension-interpolation technique, the processing block is composed of N × N pixels, the pixels are divided into object pixels and background pixels Provide a method.

To this end, the present invention provides a method comprising the steps of (A) selecting L columns of a processing block, wherein each L column in each column includes at least one object pixel; (B) determining M representing the number of object pixels included in each selected column and providing a first vector representing the values of the M object pixels; (C) selecting one tensile matrix based on M and N among a plurality of preset tensile matrices for each of the first vectors and multiplying the selected tensile matrix by the first vector to provide a tensioned first vector; (D) providing a first tensioned processing block comprising L columns, each column representing a first vector; (E) providing an L-dimensional second vector for each row of the first stretched processing block; (F) determining one tensile matrix from the plurality of predetermined tensile matrices based on the L and N values; (G) multiplying the selected tensile matrix by each second vector to provide N tensioned second processing blocks; (H) multiplying each pixel of the second stretched block by a block scaling vector to provide a tensioned block.

Description

Method and apparatus for processing an image signal including an object using a tension-interpolation technique

1 is a diagram showing one-dimensional data having a background area and an object area.

2 is a diagram showing a method of filling a background region with zeros.

3 is a diagram showing a method of filling a background area with an average value of an object area.

4 is a view showing a method for filling a background area according to a preferred embodiment of the present invention.

5 is a diagram showing a block having a background area and an object area.

6 shows a horizontally tensioned block.

7 shows a vertically tensioned block.

8 is a block diagram for explaining a video signal encoding apparatus according to a preferred embodiment of the present invention.

9 is a detailed block diagram of the tension / interpolation device of FIG.

* Explanation of symbols for main parts of the drawings

50: frame memory 100: first and coding channel

110: binary map detector 120: binary map encoder

200: block generator 300: switching circuit

400: E / I device 410: object pixel counter

420: start / sizing block 430: control block

440: RAM 460: padding block

470: tension matrix memory 480: tension block

490: rescaling block 500: second coding channel

510: transform encoder 520: quantizer

530: entropy encoder 600: formatting circuit

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a low bit rate video signal encoding apparatus and method, and more particularly, to a method and device for encoding a video signal using an extension-interpolation technique.

As is well known, in digitally broadcast systems such as video telephony, high definition television, or video conferencing systems, a significant amount of each video frame is specified because each line of video frame signal contains a series of digital data called pixels. Digital data is required. However, since the effective frequency bandwidth of conventional transmission channels is limited, in order to transmit a large amount of digital data, especially in low transmission video signal encoding systems such as video telephony and video conferencing systems, various data compression techniques are used to compress or It is necessary to reduce.

On the other hand, one of the methods for encoding an image signal in a low transmission video signal coding system is Michael Hotter, Object-Oriented Analy sis-Synthesis Coding Based on Moving Two-Dimensional Objects, Signal Processing: Image See Communication, 2, 409-428 (1990).

According to the object-specific analysis-synthesis coding method, an input image signal having moving objects is divided according to an object, and three parameters defining motion, contour, and pixel data of each object are processed through different coding paths due to their characteristics. do.

Meanwhile, in processing image data or pixels in an object, an object-based analysis-synthesis encoding technique mainly uses a transform encoding technique for removing only spatial redundancy included in the image data. One of the most commonly used transform coding techniques for image data compression is block-by-block discrete cosine transform (DCT) coding, which encodes a set of transforms for one block of digital image data, for example, 8 × 8 pixel blocks. Convert to coefficient data. This method is disclosed, for example, in Chen and Pratt, Scene Adaptive Coder, IEEE Transactions on Communications, COM-32, No. 3, pp. 225-232 (March 1984). Discrete Sine Transm, Hartley transform, or other transforms may also be used in connection with block transform encoding.

At this time, in the block-by-block DCT encoding method, a region other than the background or the object in the block is filled with 0, the average value of the pixel of the object part in the block, or a mirror image, and then converted. 2 and 3, conventional methods for filling a background area are shown for one-dimensional data. In detail, in FIG. 2, the background region is filled with zeros, and in FIG. 3, the background region is filled with average pixel values of the object region.

Although this method can use a two-dimensional DCT block used in conventional coding methods (e.g. Joint Photographic Experts Group (JPEG), Moving Pictures Experts Group (MPEG), H261, etc.), data that is not needed in the external region of the image The data compression efficiency is reduced.

An object of the present invention is to provide an encoding method capable of increasing data compression efficiency by encoding an image frame signal including an object using an extension-interpolation technique.

In order to achieve the above object, the present invention provides a method of converting a processing block included in an image frame signal including an object, wherein the processing block is composed of N × N pixels and the pixels are classified into object pixels and background pixels. Where N is a positive integer, where the object pixel is a pixel located inside the object and the background pixel is a pixel located outside the object, the method comprising: (A) selecting L columns of the processing block, each L Columns comprise at least one object pixel, and L is an integer from 0 to N; (B) For each column selected in step (A), determine M representing the number of object pixels included in each column and provide a first vector, where each element of the first vector is a respective M object A value of the pixel and M is an integer from 1 to N; (C) for each column, selecting one tensile matrix based on M and N from a plurality of predetermined tensile matrices and multiplying the selected tensile matrix by the first vector to provide a tensioned first vector; (D) providing a first tensioned processing block comprising L columns, each column comprising N pixels whose value is the element of each stretched first vector; (E) providing an L-dimensional second vector for each row of the first stretched processing block, wherein an element of the second vector is a value of a pixel included in each row of the first stretched block; (F) determining one tensile matrix from the plurality of predetermined tensile matrices based on the L and N values; (G) multiplying the tension matrix selected in step (F) by each second vector to provide N tensioned second vectors and a tensioned processing block, wherein the tensioned processing block comprises N rows and each row Comprising N pixels, the value of the pixel being an element of each tensioned second vector; (H) counting the number of object pixels in the processing block to determine a block scaling factor which is a value obtained by dividing the number of object pixels by the number of pixels in the processing block; And (I) multiplying the block scaling factor by each pixel of the second stretched processing block to provide a tensioned processing block.

Referring to FIG. 5, one block of the digital video signal is shown, where one block includes 8x8 pixels, and each pixel is represented by a rectangle. The block consists of the object area represented by the shaded pixels and the rest of the background area. The shaded pixels are called object pixels and the other pixels are called background pixels. The object pixel is stretched to fill the entire block as shown in FIGS. 6 and 7 using the extension-interpolation (E-I) technique of the present invention. To do this, as shown in Figs. 6 and 7, the horizontal and vertical tensions are performed independently of each other. Horizontal tension or vertical tension is processed before the other side, and the priority is determined according to the characteristics of the image, and horizontal tension or vertical tension is performed by line or row. If a block contains N x N pixels, for each column or row, the M-dimensional vector is converted to an N-dimensional vector, where M is a value from 1 to N. Are elements of the M-dimensional vector are M object pixels included in each column or row, and the elements of the N-dimensional vector are N stretched pixel values. For example, in the case of the third column of the block of FIG. 5, the five-dimensional vector is transformed into an eight-dimensional vector representing the third column of the vertically stretched block of FIG. 6.

At this time, the transformed M-dimensional vector f ₁ obtained by applying the M point one-dimensional DCT to the M-dimensional vector f ₁ is expressed by the following equation (1).

Wherein f ₁ (n ₁ ) is the n _1st element of f ₁ ; F ₁ (k ₁ ) is the k _1st element of F ₁ ; n ₁ and k ₁ are integers from 0 to M-1; bij is expressed as the following equation (2).

Similarly, the M-dimensional vector f ₁ is the tension using the EI method of the present invention N-dimensional vector the case of forming a f ₂ N dimensional vector transformed N-dimensional vectors to f ₂ obtained by applying the N point one-dimensional DCT f ₂ is the following It is expressed as Equation (3).

Where f ₂ (n ₂ ) is the (n ₂ ) th element of f ₂ ; F ₂ (K ₂ ) is the K _2nd element of F ₂ ; n ₂ and k ₂ are integers from 0 to N-1; Is expressed by the following equation (4).

In relation to the EI method of the present invention, the M-dimensional vector f ₁ is tensioned with the N-dimensional vector without generating additional frequency domain data. That is, the following equation (5) is satisfied.

In this case, μ ₀ is a scaling factor for equalizing the DC components of f ₁ and f ₂ , and is given by Equation (6) below.

If Equation (5) is satisfied, no additional data is generated in the frequency domain, so the EI process is optimal. From equations (1) and (3), it can be inferred that f ₂ can be obtained from f ₁ as shown in the following equations (7A) and (7B).

or

Where A and B represent NxN and NxM matrices whose members are a _ij and b _ij used in equation (7A), respectively. Equations (7A) and (7B) can be simply expressed as in the following equations (8A) and (8B).

Where C is an N × M matrix and is equal to A + B.

Using the above relationship, an object of any shape is tensioned to fill an N × N block without creating additional frequency domain elements. Conversely, the original data of FIG. 5 is recovered from the N × N block of FIG. If N is equal to M, then C is a unit matrix. Thus, the tensioning process can be omitted without changing the original vector f ₁ .

In the example shown in FIGS. 5 to 7, the third to seventh columns of the block of FIG. 5 are first stretched horizontally using the optimal method or the linear interpolation method to form the block of FIG. Similarly, the rows of the horizontally stretched block of FIG. 6 are vertically stretched using the E-I method of the present invention to form the blocks of FIG.

So far, only the case where the E-I method of the present invention is DCT has been described. However, other transforms such as Discrete Cosine Transtorm (DST), Hadamard transform, and Haar transform may be used instead. If the DST is used to encode an N × N block, the E-I method is the same as in the case of DCT except that it is represented by the following equations (9) and (10), respectively.

In the case where the image data is not highly correlated in the spatial domain, for example, in the case of inter-frame encoding in which the difference between two adjacent frames is coded, block transform coding using DST may exhibit better performance than that using DCT. It is known that there is.

In the above method, the scaling factor is applied only to the DC component of the frequency domain. However, it is sometimes advantageous to apply the scaling factor μ ₀ to other components as well. In order to do this, equation (7A) is modified as in the following equation (11).

In this case, the total amount of energy contained in the processing block simply increases during the stretching process. To reduce the effect of the energy increase, each pixel of the stretched processing block is multiplied by a block scaling factor, which is the number of object pixels in one processing block divided by N × N, the number of pixels in a block.

Referring to FIG. 8, there is shown a block diagram of a device according to a preferred embodiment of the present invention for encoding a digital video signal. The encoding apparatus of the present invention includes first and second codes, channels 100 and 500, and an extension-interpolation apparatus 400 for generating a tensioned processing block for effectively encoding a boundary of an object in a video frame. do. The first code and the channel 100 encode the contour signal of the object, and the second coded channel 500 encodes the digital video signal in block units.

A digital video signal provided from a known image source (not shown), for example a hard disk or a compact disk, is input to and stored in the frame memory 50. The digital video signal of one frame includes an object, and includes an object pixel located inside the object and a background pixel located outside thereof. Background pixels are represented by pixels with values that are much larger or smaller than the normal pixel value range. The image frame signal extracted from the frame memory 50 is provided to the binary map detector 110 of the first encoding channel 100 and the block generator 200.

Meanwhile, the first encoding channel 100 including the binary map detector 110 and the binary map encoder 120 detects a binary map of an image signal from the frame memory 50 and generates an encoded binary map by encoding by a known method. The binary map includes the same number of binary pixels as the number of pixels of the image frame signal, and each binary pixel is determined according to whether the corresponding pixel of the image frame signal is an object pixel or a background pixel. The binary map detected by the binary map detector 110 is applied to the binary map encoder 120 and encoded, and is connected to the E / I device 400 of the present invention through the line L10.

Next, the binary map encoder 120 encodes the binary map from the binary map detector 110 using, for example, a binary arithmetic code of JPEG, and supplies the encoded binary map to the formatting circuit 600.

Meanwhile, the block generator 200 divides the video frame signal from the frame memory 50 into a plurality of processing blocks having the same size including N × N pixels and provides the switching circuit 300 in blocks. In the switching circuit 300, in response to the control signal CS generated from the system controller (not shown), each processing block from the block generator 200 is selectively coupled to the E / I device 400 or the second encoding channel 500. The system controller generates the control signal CS using the contour information of the object in the video frame, which indicates whether or not the boundary of the object in the video frame exists in each processing block. If the boundary of the object is in the processing block, i.e., if the processing block includes the object area and the background area at the same time, then the processing block is coupled to the E / I device 400 which generates the tensioned processing block, otherwise the second encoding Is sent to channel 500.

According to the present invention, the E / I device 400 converts each processing block from the switching circuit 300 into a tensioned processing block in order to improve the data compression efficiency in the second encoding channel 500. In detail, the processing block as shown in FIG. 5 is supplied to the E / I apparatus 400 and converted into a tensioned processing block in the same manner as described with reference to FIG. The detailed operation of the E / I device 400 is described with reference to FIG.

Meanwhile, the second encoding channel 500 including the transform encoder 510, the quantizer 520, and the entropy encoder 530 may include image data included in each of the tensioned processing blocks from the E / I apparatus 400 using conventional transform and statistical encoding techniques. Encode the unstretched processing block from the switching circuit 300. That is, the transform encoder 510 converts the image data of the spatial domain of each processing block provided from the E / I apparatus 400 or the switching circuit 300 into a set of transform coefficients of the frequency domain by using DCT, for example, using a discrete cosine transform. A set of transform coefficients is provided to the quantizer 520. The quantizer 520 quantizes a set of transform coefficients using a known quantization method, and the set of quantized transform coefficients is provided to an entropy encoder 530 for processing.

The entropy encoder 530 also combines, for example, line length coding and variable length coding techniques to encode a set of quantized transform coefficients for each of the stretched or unstretched processing blocks from the quantizer 520 to generate an encoded video signal. do. The image signal encoded by the entropy encoder 530 is then provided to the formatting circuit 600.

And the formatting circuit 600 formats the encoded binary map from the binary map encoder 120 in the first encoding channel 100 and the encoded image frame signal from the entropy encoder 530 in the second encoding channel 500, and transmits the formatted digital image signal. In order to provide a transmitter (not shown).

Referring to FIG. 9, a detailed block diagram of the E / I apparatus 400 according to the present invention shown in FIG. 8 is shown. E / I device 400 includes object pixel counter 410, start / size determination block 420, control block 430, random access memory 440, padding block 460, tension matrix memory 470, tension block 480 and rescaling block 490. .

The processing block from the switching circuit 300 of FIG. 8 is input to and stored in the RAM 440 through the line L20. The binary map from the binary map detector 110 in FIG. 8 is input to the object pixel counter 410 and the start / size decision block 420 via line L10.

Meanwhile, the object pixel counter 410 counts the number of object pixels of each processing block in response to the binary map and provides the number of object pixels to the rescaling block 490.

In addition, in the start / size and decision block 420, the size and the start signal are determined in response to the binary map. The size signal indicates the number of object pixels in the column or row currently being processed in the processing block, and the start signal is the column being processed. Indicates the position of the first object pixel in the row. The start signal and the size signal are supplied to the padding block 460, and the size signal is connected to the tension matrix memory 470 and the tension block 480.

Next, the detailed functions of the above start / size determination block 420 will be described with reference to the examples of FIGS. If there are no object pixels in the current column, such as the first and second columns of the processing block of FIG. 5, the size signal is not required for the padding block 460, the tension matrix memory 470, and the tension block 480. Inform them. In the case of the third column of the processing block of FIG. 5, the start signal indicates that the third pixel is the first object pixel, and the size signal indicates that there are five object pixels in the column. At this time, during the vertical tensioning, the start signal and the size signal are fixed to 3 and 6, respectively, because the starting position and the number of object pixels in each row of the horizontally tensioned block are the same as shown in FIG.

Meanwhile, the tension matrix memory 470 stores a tension matrix for converting an M-dimensional vector into an N-dimensional vector, that is, C of Equation (8B). N is determined by the system design, and in many cases eight. Accordingly, it is possible to pre-calculate the tension matrix C for all M values, namely 1 to N, and store it in the tension matrix memory 470, and thus the tension matrix memory 470 may be configured as a read only memory (ROM).

Tensile-interpolation for a column (or row) of such a block can be easily performed by multiplying the appropriate tensile matrix stored in the tension matrix memory 470 of the M-dimensional vector formed with the values of the object pixels of that column (or row). The appropriate tension matrix is selected in response to the size signal and provided to tension block 480 for multiplication. If the size signal indicates that there are no object pixels in the current column or row, then no tension matrix is selected or provided to tension block 480.

In the control block 430, an H / V signal and a column / row number signal are generated. The H / V signal indicates whether horizontal or vertical tension is being performed in the E / I apparatus of the present invention. The / row number signal indicates a column or row currently being processed in the E / I device 400 of the present invention. In general, the columns / rows of a block are processed in order, so the signal is determined by the system design, for example the number of columns and rows of processing blocks. The H / V signal and the column / row signal are provided to the start / size decision block 420 and the RAM 440.

As described above, the processing block from the switching circuit 300 of FIG. 8 is provided to the RAM 440. The columns or rows of processing blocks are sequentially supplied from the RAM 440 in response to the H / V signals and column / row number signals from the control block. For example, it is assumed that horizontal tensioning is performed first. In this case each row of processing blocks is first provided to padding block 460.

Next, in padding block 460, in response to the start signal and the size signal, an M-dimensional vector representing object pixel values included in columns provided from RAM 440 is formed and provided to tension block 480. The M-dimensional vector can be created by selecting M consecutive object pixels from the position indicated by the start signal.

In addition, one tension matrix is selected from a plurality of preset tension matrices stored in the tension matrix memory 470 in response to the size signal, that is, the M value signal, and provided to the tension block 480. The tension matrix is multiplied by the M-dimensional vectors in tension block 480 to form a tensioned N-dimensional vector. N-dimensional vectors or horizontally stretched columns are again stored in the location where the original columns of RAM 440 were stored.

In the same manner as described above, one row of processing blocks stored in RAM 440 is converted into horizontally stretched rows and stored again. All rows of the processing block are processed in a similar manner except for columns that do not contain object pixels. For columns that do not contain object pixels, the size signal indicates that there is no object pixel in tension block 480 so that multiplication is not performed and the original data in RAM 440 is maintained.

Thus, when the horizontal pulling is finished, it can be seen that the horizontally tensioned processing block containing the stretched rows as shown in FIG. 6 is stored in the RAM 440.

Then, after the horizontal stretching is finished, the rows of the horizontally stretched processing block are sequentially input from the RAM 440 to the padding block 460 in response to the H / V signal and the column / row number signal from the control block 430. Except for this, the functions of the padding block 460, the tension matrix memory 470, and the tension block 480 are the same as those of the horizontal tension. Specifically, M-dimensional vectors formed from each row of horizontally tensioned processing blocks are provided from padding block 460 to tension block 480 and converted there into N-dimensional vectors. The tensioned row (M or N-dimensional vector) is stored at the location where the corresponding row of the horizontally stretched processing block of RAM 440 was.

After the horizontal and vertical tensioning ends, the tensioned processing blocks stored in RAM 440 are provided to rescaling block 490. In rescaling block 490, each pixel value of the tensioned processing block is scaled down using a block scaling factor, which in response to the number of object pixels in one processing block provided from object pixel counter 410. Is determined.

The rescaled stretched processing block is then connected to and encoded therein with a transform encoder 510 of the second encoding channel 500 as shown in FIG.

As described above, according to the method of the present invention, the coding efficiency can be improved by significantly reducing the high frequency components existing between the pixels inside the object and the pixels outside the object.

Claims (5)

A method of converting a processing block included in an image frame signal including an object, wherein the processing block is composed of N × N pixels, the pixels are classified into an object pixel and a background pixel, and N is a positive integer. The pixel located inside the object and the background pixel being the pixel located outside of the object; (A) selecting L columns of the processing block, each L columns comprising at least one object pixel, where L is an integer from 0 to N; (B) For each column selected in step (A), determine M representing the number of object pixels contained in each column, and provide a first vector, where each element of the first vector is each M A value of the object pixel, wherein M is an integer from 1 to N; (C) for each column, selecting one tensile matrix based on M and N from a plurality of predetermined tensile matrices and multiplying the selected tensile matrix by the first vector to provide a tensioned first vector; (D) providing a first tensioned processing block comprising L columns, each column comprising N pixels whose value is the element of each stretched first vector; (E) providing an L-dimensional second vector for each row of the first stretched processing block, wherein an element of the second vector is a value of a pixel included in each row of the first stretched block; (F) determining one tensile matrix from the plurality of predetermined tensile matrices based on the L and N values; (G) multiplying the tensile matrix selected in step (F) by each second vector to provide N tensioned second vectors and a tensioned processing block, wherein the tensioned processing block comprises N rows, each row Comprises N pixels, wherein the value of the pixel is an element of each tensioned second vector; (H) counting the number of object pixels in the processing block to determine a block scaling factor that is a value obtained by dividing the number of object pixels by the number of pixels in the processing block; And (I) multiplying the block scaling factor by each pixel of the second stretched processing block to provide a tensioned processing block. The method of claim 1 wherein the tension matrix for converting the M-dimensional vector to an N-dimensional vector Is determined as, where a _ij is Where b _ij is Expressed as, μ ₀ is The coding method represented by. The method of claim 1, wherein the tension matrix for converting the M-dimensional vector to an N-dimensional vector Is determined as follows, where b _ij is Represented by a _ij Expressed as, μ ₀ is The coding method represented by. A method of converting a processing block included in an image frame signal including an object, wherein the processing block is composed of N × N pixels, the pixels are classified into object pixels and background pixels, N is a positive integer, and an object pixel is an object. A pixel located inside of the pixel, and the background pixel is a pixel located outside of the object; (A) selecting L columns of the processing block, each L columns comprising at least one object pixel, where L is an integer from 0 to N; (B) For each column selected in step (A), determine M representing the number of object pixels contained in each column and provide a first vector, where each element of the first vector is a respective M object pixel. Wherein M is an integer from 1 to N; (C) for each column, selecting one tensile matrix based on M and N from a plurality of predetermined tensile matrices and multiplying the selected tensile matrix by the first vector to provide a tensioned first vector; (D) providing a first tensioned processing block comprising L columns, each column comprising N pixels whose value is the element of each stretched first vector; (E) for each row of the first stretched processing block, providing an L-dimensional second vector, wherein an element of the second vector is a value of a pixel included in each row of the first stretched block; (F) determining one tensile matrix from the plurality of predetermined tensile matrices based on the L and N values; (G) multiplying the tension matrix selected in step (F) by each second vector to provide N tensioned second vectors and a tensioned processing block, wherein the tensioned processing block comprises N rows and each row A method comprising N pixels, wherein a value of the pixel is an element of each tensioned second vector, wherein the tension matrix for converting the M-dimensional vector into an N-dimensional vector includes: Is determined as, where aij Represented by bij, Where μ ₀ is The coding method represented by. An apparatus for encoding an image frame signal including an object, wherein the image frame signal includes an object pixel and a background pixel, wherein the object pixel is a pixel located inside the object and the background pixel is a pixel located outside the object. Is; (A) means for detecting a binary map indicating which of the pixels of the image frame signal is an object pixel; (B) means for encoding the binary map to generate a first encoded image frame signal; (C) means for dividing an image frame into a plurality of processing blocks, each processing block comprising N × N pixels, where N is a positive integer; (D) means for generating a control signal indicating whether each processing block includes an object pixel and a background pixel at the same time; (E) means for classifying a processing block into a first set of processing blocks and a second set of processing blocks in response to a control signal, wherein each processing block of the first set includes both an object pixel and a background pixel; Way; (F) conversion means for converting each processing block of the first set into a tensioned processing block; (G) means for encoding the stretched processing block or the second set of processing blocks to provide a second coded image frame signal; (H) an apparatus comprising means for formatting first and second encoded video frame signals, the means for transforming each processing block; (F1) means for selecting L columns of the processing block, each L columns comprising at least one object pixel, wherein L is first selection means that is an integer from 0 to N; (F2) For each column selected by the first selection means, determine M representing the number of object pixels contained in each column and provide a first vector, wherein each element of the first vector is each M object Means for pixels, wherein M is an integer from 1 to N; (F3) means for selecting one tensile matrix based on M and N among a plurality of predetermined tensile matrices for each of the columns, and multiplying the selected tensile matrix by the first vector to provide a tensioned first vector, wherein Tensile matrix for converting M-dimensional vectors into N-dimensional vectors. Is determined as, where aij is Represented by bij, Is expressed as μ ₀ , Represented by; (F4) means for providing a first tensioned processing block comprising L columns, each column comprising N pixels whose value is an element of each tensioned first vector; (F5) means for providing an L-dimensional second vector for each row of the first stretched processing block, wherein an element of the second vector is a value of a pixel included in each row of the first stretched block; (F6) second selection means for determining one tensile matrix from a plurality of predetermined tensile matrices based on L and N values; (F7) means for multiplying each second vector by the tensile matrix selected in the second selection means to provide N tensioned second vectors and a tensioned processing block, the tensioned processing block comprising N rows and each And the row comprises N pixels, wherein the value of the pixel comprises means that is an element of each tensioned second vector.