CN101184227A - Block removing filter - Google Patents

Abstract

A block-removing filter comprises two sets of filter calculation circuits, a transposition buffer for converting the data filtering direction, an output buffer area for storing intermediate values and final results in the calculation process, and a control and output module for judging various different condition combinations and being responsible for processing external input and output data. The filter for removing block in the invention is a filter structure suitable for pure frame picture or pure field picture coding, or frame and field picture coding macro block adjustable, and can effectively shorten the filtering time and reduce the circuit area by the configuration of internal buffer and the control of filtering flow.

Description

deblocking filter

technical field

The present invention relates to an image filter architecture, in particular to an H.264 main-level and high-level filter architecture.

Background technique

In the current newer video compression technologies H.264 and MPEG4, a de-blocking filter (De-blocking filter) is added, its function is to improve the compression rate of the encoder, and after decoding, the picture is removed Block smoothing and filtering can effectively improve the quality of the output image.

According to the H.264/AVC standard, when the filter performs filtering, four pixels (pixels) are extracted from the left and right sides of the boundary to be filtered, and the filtering is performed. As shown in FIG. 1 , a schematic diagram of the filtering operation of the known deblocking filter, the filtering sequence must first complete the horizontal filtering, and then perform the vertical filtering. If a giant block is divided into 16 small blocks, in terms of a 4×4 (pixel) small block, four times of horizontal filtering and four times of vertical filtering are required to complete the entire filtering action.

In order to complete the above actions and take into account the speed requirements, in some existing designs, the main part of the discussion is mostly through the design of the filtering sequence, so that the vertical and horizontal filtering actions can be performed simultaneously; secondly, due to the When performing filtering, each 4×4 small block must undergo four filtering operations, so some register buffers must be set up to store the blocks that only partially complete the filtering operation. Therefore, if the complete filtering action of a block can be completed faster in the design, the less buffer space is needed and the design area is smaller, and this requirement can also be achieved through a good filtering sequence.

Please refer to FIG. 2 , which is a schematic diagram of a known filtering sequence of a deblocking filter, which is a relatively common filtering sequence in current existing designs. The largest block above is a giant block of Luma, and the smaller one below is the corresponding giant block of Chroma. A 4×4 small block is used as the cutting unit, and each small block has four boundaries to be filtered, and the numbers on each boundary represent the order of filtering. Since the rule of doing the horizontal direction first, then the vertical direction, it can be seen from the small block in the upper left corner of the picture that the first boundary in the vertical direction must wait until the two boundaries in the horizontal direction are filtered. Get started. The same number on the boundary means that the filtering time is also the same.

However, the known architecture of the deblocking filter can only handle macroblocks that only contain Frame or Filed codes in one picture, and are designed for the main level of H.264; but in H. 264 high-level, the frame and field codes of the macroblocks in the same picture can be changed with each macroblock pair (Macroblock pair), so the known structure of the deblocking filter cannot be applied. Since each giant block is performing filtering, 4 pixels of the border part must be taken out from the top and left giant blocks at the same time. Therefore, if the frame and field coding will change with the giant block, when performing the filtering action at the junction, under different boundary conditions, the filtering process will be different from the pure frame image or pure field image encoding, so the design must be Consider these different boundary conditions.

Contents of the invention

In view of this, the present invention provides a filter architecture applicable to a pure frame image or a pure field image, or a macroblock-adaptive frame-field coding (Macroblock-Adaptive Frame-Field Coding). Moreover, the present invention can achieve the effects of reducing the circuit area and shortening the filtering time through the design of the filtering structure and the control of the filtering process.

For reaching above-mentioned purpose, the present invention provides a kind of deblocking filter, it is characterized in that, comprises a horizontal filter, is to be used for receiving a giant block, and this giant block is carried out the filtering of horizontal direction; A vertical filter, It is used to filter the giant block in the vertical direction; an output buffer is connected between the horizontal filter and the vertical filter, and is used to temporarily store the giant block filtered in the horizontal direction and filtered in the vertical direction After the giant block; a transposition buffer is connected to the horizontal filter, the vertical filter and the output buffer, and receives the small block of the giant block after the horizontal filter has been done, and is used to convert the The small blocks of the giant block after horizontal filtering are transposed for the vertical filter to perform vertical filtering; a data management unit is connected to the output buffer to judge the encoding state of the mega block, and Outputting the giant block after horizontal filtering and vertical filtering; and a control unit connected to the horizontal filter, the vertical filter, the transposition buffer and the data management unit for controlling the horizontal filter, The vertical filter, the transpose buffer and the data management unit perform filtering operations on the megablock.

The present invention also proposes a method for removing block filtering, which is characterized in that at first reading a first giant block required for filtering; then performing filtering of a first giant block and reading a second block simultaneously The upper giant block required by the giant block; then carry out the filtering of a second giant block, and write back the filtering result of the first giant block in a dynamic random access memory at the same time; finally write the filtering result of the second giant block back into the DRAM.

Description of drawings

FIG. 1 is a schematic diagram of a filtering action of a known deblocking filter;

Fig. 2 is a schematic diagram of the filtering sequence of a known deblocking filter;

Fig. 3 is a flow chart of data control for the deblocking filter architecture of the present invention;

Fig. 4 is a system diagram for removing the blocking filter of the present invention;

Fig. 5 is a configuration diagram for the output buffer of the present invention;

Fig. 6 is a configuration diagram for the transposition buffer of the present invention;

Fig. 7 is a schematic diagram of the filtering sequence for the current giant block in the block wave filter of the present invention as frame image encoding and the upper giant block as field image encoding

FIG. 8 is a schematic diagram of the filtering sequence under normal conditions of the deblocking filter of the present invention.

Symbol Description

deblocking filter 1

control unit 101

input buffer 102

Horizontal filter 103

vertical filter 104

transpose buffer 105

Buffer 1051

output buffer 106

The first scratch buffer 1061

Second scratch buffer 1062

Buffer array 1063

Data Management Unit 107

Data processing unit 108

Operation comparison unit 109

Direct Memory Access Unit 110

DRAM 2

Small blocks A, B, C, D, E, F, G, H, I, J, K, L, M, N, O, P, Q, R, S, T, U, V, W, X

Detailed ways

Since a group of giant squares includes a luma giant square (Y) and two chrominance giant squares (Cb, Cr), and when filtering each giant square, it is necessary to take out the boundary part from the upper and left giant squares of 4 pixels. When the image filter architecture of the present invention performs the filtering of the current giant block, it needs to read the data of the upper block from the dynamic random access memory, and temporarily store the data required by the previous block for the filtering of the current block use.

Please refer to Fig. 3 the flow chart of data control of the block filter architecture of the present invention, the steps of the overall data control flow of the image filter architecture of the present invention are at first after the current giant block is ready (referring to the current giant block in the left giant block) When the filtered data has been stored in a register), it starts to read the upper giant block pixels required by the luma block for filtering in the DRAM (step S301 in FIG. 3 ). After reading the required upper giant square of the brightness square when filtering, the filtering of the brightness square is carried out, and at the same time, the upper giant square pixels required by the chroma giant square are read in the dynamic random access memory (step S303 in FIG. 3 . ). Then, after completing the filtering of the luma block and reading the pixels of the upper giant block required by the chroma block, the filtering of the chroma block is started, and at the same time, the filtered result of the luma block is written back to the dynamic random access memory (as shown in Figure 3 Step S305). Finally, after the filtering of the chrominance giant block is completed, the result of the chrominance giant block is written back into the dynamic random access memory (see step S307 in FIG. 3 ).

The above flow control can simultaneously read the chroma block and write back the luminance block to the DRAM while performing luma and chroma filtering. Because the access of dynamic random access memory is not timely, after sending out the request (Request), it needs to be queued with other modules, and can only get a reply after a period of time. Therefore, the main reason for doing this is to shorten the time required for the entire filtering process by hiding the delay time of the DRAM in the filtering process.

Next, please refer to FIG. 4 , which is a system diagram of the deblocking filter of the present invention. The deblocking filter 1 of the present invention includes a control unit 101, an input buffer 102, a horizontal filter 103, a vertical filter 104, a transpose buffer 105, an output buffer 106, a data management unit 107, a data processing unit 108, An operation comparison unit 109 and a direct memory access unit 110 . The direct memory access unit 110 generates the address of the DRAM 2 for the data management unit 107 to write data into the DRAM 2 . The input register 102 is used for temporarily storing macroblocks to be read from the DRAM 2 , including luma macroblocks and chrominance macroblocks. When the luma giant block and the chroma macro block are ready, they are provided to the horizontal filter 103 to start filtering in the horizontal direction, and when performing luma and chroma filtering, the chroma block is simultaneously performed on the DRAM 2 to read. The data processing unit 108 is used to detect the encoding status of adjacent giant blocks, such as detecting that the adjacent giant blocks are Frame or field encoded giant blocks, or that the frame and field encoding can be Changed macroblock pair (Macroblock pair), etc.; and the operation comparison unit 109 is used for the processing of the macroblock parameters, such as the comparison of Alpha, Beta and bs values, and then allows the control unit 101 to make combinations of various situations out the corresponding filter control.

In addition to blocking filter 1, in order to perform horizontal and vertical filtering operations at the same time, two sets of filter (Filter) calculation circuits are used in the architecture, horizontal filter 103 and vertical filter 104 for filtering; transpose buffer District 105 (Transpose Buffer) is used for converting data filtering direction; Stores the intermediate value and the final result in the calculation process by output buffer 106; To judge various combinations of different situations and be responsible for processing external input and output data.

The deblocking filter 1 of the present invention performs filtering in the horizontal direction and vertical direction at the same time, and two sets of calculation circuits are used in the design. Compared with the design of completing the horizontal filtering first and then performing the vertical filtering, it only costs 1/ The time of 2 can complete the filtering of a group of giant blocks.

In addition, in order to avoid unnecessary access to the dynamic random access memory 2, to improve the utilization efficiency of the dynamic random access memory 2, when the control unit 101 performs the judgment before filtering, if the giant block to be filtered at present, the uppermost Or the far left does not need to be filtered, then the actions of reading the brightness square and the chroma square will be skipped during filtering, and the data of the giant square above will not be read from the dynamic random access memory 2; After the filtering of one group of giant blocks, the data to be filtered by the next group of giant blocks is directly stored in the output buffer 106, so that when performing the filtering of the next group of giant blocks, there is no need to use dynamic random access again. Memory 2 reads the data of the giant square on the left. The two designs mentioned above can also significantly shorten the average time of filtering.

Please refer to FIG. 5 which is a configuration diagram of the output buffer of the present invention. The configuration of the output buffer 106 includes a first temporary buffer 1061 , a second temporary buffer 1062 and a register array 1063 (register array). The upper register array 1063 is used to store the upper giant square pixels required for the uppermost border filtering, and the second temporary storage buffer 1062 on the left is used to store the adjacent giant square pixels required for the leftmost border filtering. pixels. The number of adjacent macroblocks that needs to be stored is only 4×16×8 because there is no consideration of macroblock-adaptive frame-field coding (Macroblock-Adaptive Frame-Field Coding) at the H264 main level (bit) temporary storage buffer, and the way of reading is relatively simple. In the high-level H264, if the frame and field image encoding megablocks are adjustable, the transmitted megablocks are in pairs, so the number to be stored becomes twice the original 8×16×8 (bits), The access method will also change with the encoding methods of the current giant block and the adjacent giant block. The first temporary buffer 1061 is the output buffer of the current giant block. The first temporary buffer 1061 and the second temporary buffer 1062 are a static random access memory.

Please refer to FIG. 6 which is a configuration diagram of the transposition buffer of the present invention. During the calculation process of deblocking filter 1, the block value filtered in the horizontal direction for the first time will be stored in the first temporary storage buffer 1061, and then taken out for the second horizontal filtering action, and the completed After the value is stored in the transpose buffer 105. The transposition buffer 105 is composed of seven groups of 4×4×8 (bit) registers 1051, and each group of 4×4×8 registers 1051 can be used to store a small block that has been filtered in the horizontal direction. And the filtering in the other direction is performed by reading in the vertical direction. After the filtering in the vertical direction is also completed, write back to the first temporary buffer 1061 at last. The right side of FIG. 6 shows the configuration structure diagram of the 4×4×8 buffer 1051. When the small block for horizontal filtering is completed, the completed small block is input from the first input terminal and then output from the first output terminal. for vertical filtering. And when the filtering in the vertical direction is completed, it is input from the second input terminal and read out from the second output terminal, so as to write back to the first temporary buffer 1061 correctly. This approach is used to reduce the number of registers to reduce the area of the circuit. And when all the small blocks in the first temporary storage buffer 1061 have been filtered, the processed giant block data can be written into the DRAM 2 through the data management unit 107, and this action can be used as When the chroma block is filtering, it writes the filtered luma block back into the DRAM 2 at the same time, as described in the above control process.

The input and output of each buffer 1051 mentioned above are only input and output from a fixed buffer, which reduces the complexity of wiring and circuit area, and when the frame and field image coding giant blocks are adjustable, data reading and writing The method has also been changed and designed accordingly, so that the entire architecture can operate smoothly when the frame and field image encoding giant blocks are adjustable under normal conditions, without wasting additional buffer and filtering time.

Therefore, the structure of the output buffer 106 and the transposition buffer 105 is applicable to both the main level and the high level of H264, but the data reading and writing method is more complicated at the high level, and will be based on the encoding of the current megablock and adjacent megablocks There are different ways.

The data access action of the output buffer 106 is different from the general situation when the current giant block is frame image encoding and the upper giant block is field image encoding. As shown in FIG. 5 , in order to reduce the complexity of wiring and control, each set of register arrays 1063 used to store small blocks has its input and output from the same set of registers in all directions, so it cannot be arbitrarily designated. Any of the four registers is used to access the data therein. Under normal circumstances, the data access method of the block filter 1 is based on the encoding method of the current giant block, and the current giant block is field image encoding, so the data access method during filtering is field image access. And vice versa. Only in the special situation described above, although the current giant block is coded as a frame image, when filtering with the giant block of the upper field map, the data must be accessed in the form of a field map. After the two borders are filtered, switch back to frame image access.

Please refer to FIG. 7 , which is a schematic diagram of the filtering sequence in which the current giant block is frame image encoding and the upper giant block is field image encoding in the deblocking wave device of the present invention; Schematic diagram of the filtering sequence. For various boundary conditions in the design, the order of filtering is changed as shown in Figure 7 and Figure 8. In Figure 7, when the current giant block is coded as a frame image and the giant block above is coded as a field image, to make the first boundary in the vertical direction on the left, the pixels in the two small blocks above must be used. This is because when the upper field image is coded, four pixels of the top field or bottom field must be taken out from the giant block of the current frame image code when performing the filtering action, so the filtering time in the vertical direction must be delayed , it cannot start until the last two small blocks have finished filtering in the horizontal direction. Fig. 8 is a schematic diagram of the filtering order of the block filter of the present invention under normal conditions. It also obeys the rule that the horizontal direction is performed first, and then the vertical direction is used for filtering, but the filtering direction is the first one in the vertical direction on the left. The boundary starts to work, which is different from the direction of general filtering in the past. Secondly, because the output buffer 106 also has a temporary storage area for intermediate values during filtering, in order to prevent the first temporary buffer 1061 from reading and writing the same address at the same time, the direction of filtering is changed.

In order to further explain the direction of filtering and the filtering action of the giant block, please refer to FIG. 4, which is a system diagram of the deblocking filter of the present invention, and FIG. The big square above is a schematic diagram of the filtering sequence of field image coding. When the current giant block in the deblocking filter of the present invention is frame-image coded, and the upper block and block are field-map coded, the upper luminance giant block is first made and then the lower two chrominance giant blocks are made. Its filtering action is that initially the horizontal filter 103 performs the horizontal filtering of the first side on the small blocks in the vertical direction of the luminance macroblock in sequence (as shown in Figure 7 numbered 1, 2, 3, 4), and after the first side is executed After horizontal filtering of , it is written into the output buffer 106 . Then the horizontal filter 103 reads out the small blocks in the vertical direction of the luminance square in the output buffer 106 in sequence, and performs horizontal filtering on the second side (as shown in Figure 7 numbered 5, 6, 7, 8), and after executing the second After the horizontal filtering of the two sides, it is written into the transposition buffer 105 . And because when the upper part is the field image encoding, when performing the filtering action, it is necessary to take out four pixels of the top field or the bottom field from the giant block of the current frame image encoding, so the filtering time in the vertical direction of the small block A (such as No. 8) in FIG. 7 must be delayed, so that the vertical filtering of the small block A is executed after the horizontal filtering of the small block A and the small block B (a small block pair) is completed. And the same is true for the small blocks E, I, M, and the small blocks Q, S, U, W of the chroma square. When performing the vertical filtering of the small block A, read out the small block A that has completed the horizontal filtering of the first side and the second side in the transposition buffer 105 (such as the small block A with the filter numbers 1 and 5 completed in FIG. 7 ) to perform the vertical filtering of the first side of the small block A (as shown in FIG. Then read out the small block B that has finished the horizontal filtering of the first side and the second side in the transposition buffer 105 (such as the small block B that has finished numbering 2 and 6 filtering in Figure 7), to execute the numbering of the small block B 9 vertical filters. Temporarily store the small blocks that have been filtered horizontally and vertically in the output buffer 106 (such as the small blocks A that have been filtered with numbers 1, 5, 8, and 9 in FIG. 7 ). Finally, after all the small blocks (block A~block P) in the luminance macroblock have been filtered, the entire luminance macroblock is written back into the DRAM 2 from the output buffer 106 . In the same way, the filtering of the chroma megablock is performed.

Next, please refer to FIG. 4 , a system diagram of the deblocking filter of the present invention, and FIG. 8 , a schematic diagram of the filtering sequence of the deblocking filter of the present invention under normal conditions. When the present invention removes the block filter under normal conditions (pure frame image or pure field image) for filtering, the upper luminance giant block is made first and then the lower two chrominance giant blocks are made. At the beginning, the horizontal filter 103 sequentially performs the horizontal filtering of the first side on the small blocks in the vertical direction of the luminance macroblock (as shown in the numbers 1, 2, 3, and 4 in Figure 8), and performs the horizontal filtering of the first side After that, it is written into the output buffer 106. Then the horizontal filter 103 reads out the small blocks in the vertical direction of the luminance square in the output buffer 106 in sequence, and performs the horizontal filtering of the second side (as shown in Figure 8, No. 5, 6, 7, 8), and after the execution of the second After the horizontal filtering of the two sides, it is written into the transposition buffer 105 . When the horizontal filtering is executed to number 7, the small block A that has completed the horizontal filtering of the first side and the second side is read in the transposition buffer 105 (such as the small block A that has completed the filtering of numbers 1 and 5 in Figure 8) , to perform the vertical filtering of the first side of the small block A (as shown in FIG. 8 , the filtering of the small block A number 7 is performed), and then temporarily store the small block A in the transposition buffer 105 . Then do the vertical filtering of the small block B number 8 in the same way. Temporarily store the small block after horizontal and vertical filtering in the output buffer 106 (such as small block A with numbers 1, 5, 7, and 8 filtered in FIG. 8 ). Finally, after all the small blocks (block A~block P) in the luminance macroblock have been filtered, the entire luminance macroblock is written back into the DRAM 2 from the output buffer 106 . In the same way, the filtering of the chroma megablock is performed.

However, the above descriptions are detailed descriptions and accompanying drawings of preferred specific embodiments of the present invention. The features of the present invention are not limited thereto, and are not intended to limit the present invention. All scopes of the present invention should be based on the claims , all embodiments that conform to the spirit of the claims of the present invention and similar changes thereof shall be included in the scope of the present invention, and any change or modification that can be easily conceived by those skilled in the art in the field of the present invention is all possible fall within the scope of the claims in this case.

Claims (18)

1. a deblocking filter is characterized in that, comprising:

One horizontal filter is to be used to receive a huge square, and this huge square is carried out the filtering of horizontal direction;

One vertical filter is to be used for this huge square is carried out the filtering of vertical direction;

One output buffer is to be connected between this horizontal filter and this vertical filter, is used for temporary through filtered this huge square of horizontal direction and through filtered this huge square of vertical direction;

One transpose buffering district, be to be connected in this horizontal filter, this vertical filter and this output buffer, receive the block of cells of having finished filtered this huge square of horizontal direction, the block of cells transposition that is used for finishing filtered this huge square of horizontal direction carries out the filtering of vertical direction for this vertical filter;

One Data Management Unit is to be connected in output state, in order to judge the encoding state of this huge square, reaches this huge square that horizontal filtering and vertical filtering are finished in output; And

One control unit, be to be connected in this horizontal filter, this vertical filter, this transpose buffering district and this Data Management Unit, in order to control this horizontal filter, this vertical filter, this transpose buffering district and the execution of this Data Management Unit filtering action to this huge square.

2. deblocking filter as claimed in claim 1 is characterized in that this huge square is to be huge square of a colourity or the huge square of a brightness.

3. deblocking filter as claimed in claim 1 is characterized in that comprising an input buffer, is to be connected in this horizontal filter, be used to store will carry out filtering this huge square so that this horizontal filter filtering to be provided.

4. deblocking filter as claimed in claim 3 is characterized in that this huge square is to comprise huge square of a colourity and the huge square of a brightness.

5. deblocking filter as claimed in claim 1 is characterized in that comprising a computing comparing unit, is acquisition and the comparison that is used for this huge square parameter, judges for this control unit whether this present huge square needs to carry out filtering.

6. deblocking filter as claimed in claim 1 is characterized in that comprising a data processing unit, is to be connected in this information management unit, is used to detect adjacent and the encoding state of an adjacent huge square of this huge square at present.

7. deblocking filter as claimed in claim 1, it is characterized in that comprising a direct memory access unit, be to be connected between this Data Management Unit and a DRAM (Dynamic Random Access Memory), be used to produce the address of this lively attitude random access memory, for this Data Management Unit with this huge square of finishing filtering in this output buffer this DRAM (Dynamic Random Access Memory) of writing direct.

8. deblocking filter as claimed in claim 1 is characterized in that this output buffer is to comprise:

One first temporary storage buffer region is to be used for storing this huge square;

One second temporary storage buffer region is to be used to store an adjacent huge square required when carrying out this huge square Far Left boundary filtering; And

One buffer array is to be used for storing carrying out this huge square topmost during boundary filtering, the required huge square of using in a top.

9. deblocking filter as claimed in claim 8 is characterized in that this second temporary storage buffer region is the buffer for one 8 * 16 * 8.

10. deblocking filter as claimed in claim 8 is characterized in that this buffer array all is to carry out from same group of buffer in the input and output of all directions.

11. deblocking filter as claimed in claim 1 is characterized in that this transpose buffering district comprises many group buffers, with the block of cells of temporary this huge square.

12. deblocking filter as claimed in claim 11 is characterized in that those buffers are respectively is one 4 * 4 * 8 buffer.

13. one kind is removed the blocking filtering method, it is characterized in that step comprises:

The required huge square in a top when reading one first huge square filtering;

Carry out the filtering of one first huge square, read the required huge square in top of one second huge square simultaneously;

Carry out the filtering of one second huge square, the result with this first huge square filtering writes back in the DRAM (Dynamic Random Access Memory) simultaneously; And

The filtering result of this second huge square is write back in this DRAM (Dynamic Random Access Memory).

14. the blocking filtering method of removing as claimed in claim 13 is characterized in that the first huge square is to be the huge square of a brightness, the second huge square is to be the huge square of a colourity.

15. the blocking filtering method of removing as claimed in claim 14 is characterized in that the filter step of this first huge square normal condition comprises:

In regular turn a block of cells of this first huge square vertical direction is carried out the horizontal filtering on first limit, and behind the horizontal filtering that executes first limit, write in the output state;

In regular turn this block of cells of this first huge square vertical direction is carried out the horizontal filtering on second limit, and behind the horizontal filtering that executes second limit, write in the transposition buffer;

In this transposition working area, read this block of cells of finishing first limit and the second limit horizontal filtering, finish the first limit vertical filtering of this block of cells of first limit and the second limit horizontal filtering with execution;

Be temporary in this transposition buffer with this block of cells of finishing the first limit vertical filtering finishing first limit and the second limit horizontal filtering;

In this output state, read next block of cells of finishing first limit and the second limit horizontal filtering, to carry out the first limit vertical filtering of this next block of cells;

This block of cells of finishing level and vertical filtering is temporary in this output working area; And

After all block of cells are all finished filtering in this first huge square, just write back in this DRAM (Dynamic Random Access Memory).

16. the blocking filtering method of removing as claimed in claim 15 is characterized in that the filter step of this second huge square normal condition comprises:

In regular turn a block of cells of this second huge square vertical direction is carried out the horizontal filtering on first limit, and behind the horizontal filtering that executes first limit, write in the output state;

In regular turn this block of cells of this second huge square vertical direction is carried out the horizontal filtering on second limit, and behind the horizontal filtering that executes second limit, write in the transposition buffer;

In this transposition working area, read this block of cells of finishing first limit and the second limit horizontal filtering, finish the first limit vertical filtering of this block of cells of first limit and the second limit horizontal filtering with execution;

Be temporary in this transposition buffer with this block of cells of finishing the first limit vertical filtering finishing first limit and the second limit horizontal filtering;

In this output state, read next block of cells of finishing first limit and the second limit horizontal filtering, to carry out the first limit vertical filtering of this next block of cells;

This block of cells of finishing level and vertical filtering is temporary in this output working area; And

After all block of cells are all finished filtering in this second huge square, just write back in this DRAM (Dynamic Random Access Memory).

17. the blocking filtering method of removing as claimed in claim 14, it is characterized in that this first huge square is that hardwood graph code, the huge square in a top are the filtering of field pattern coding, the filtering time of the top block of cells vertical direction of this first huge square must postpone backward, and the vertical filtering that makes the top block of cells of this first huge square is to wait until after this top block of cells and this next block of top block of cells are all finished the filtering of horizontal direction to begin to carry out again.

18. the blocking filtering method of removing as claimed in claim 17, it is characterized in that this second huge square is that hardwood graph code, the huge square in this top are the filtering of field pattern coding, the filtering time of the top block of cells vertical direction of this second huge square must postpone backward, and the vertical filtering that makes the top block of cells of this first huge square is to wait until after this top block of cells and this next block of top block of cells are all finished the filtering of horizontal direction to begin to carry out again.

Images