TWI290438B - A pipelined deblocking filter - Google Patents

Abstract

An apparatus and method for pipelined deblocking includes a filter having a filtering engine, a plurality of registers in signal communication with the filtering engine, a pipeline control unit in signal communication with the filtering engine, and a finite state machine in signal communication with the pipeline control unit; and a method of filtering a block of pixel data processed with block transformations to reduce blocking artifacts includes filtering a first edge of the block, and filtering a third edge of the block no more than three edges after filtering the first edge, wherein the third edge is perpendicular to the first edge.

Description

1290 pifdoc IX, invention description: • [Technical field of invention] The contents of the transcript are related to video encoders and decoders (collectively referred to as "CODEC"), and in particular with deblocking filters ( Deblocking filter ) image C0DEC. A pipelined filtering apparatus and method for avoiding blocking artifacts is provided. [Prior Art] Image poor materials are generally processed and transported in the form of bitstreams. Image encoders typically use a block transform coding [such as discrete cosine transform (DCT)] to compress the original data. The corresponding video decoder typically decodes the bitstream data that has been subjected to block transform coding, for example by applying an inverse discrete cosine transform (IDCT), to decompress the block. Digital image compression technology converts natural imagery into a compressed image without significant loss of quality. Many image compression standards have emerged, including H.26, H.263, MPEG_b MPEG-2, and MPEG-4. Compared with the compression standard of the first rain, the proposed ITU-T recommended H.264 ISO/IEC 14496-10 AVC image compression standard (H.264/AVC) has achieved coding efficiency in the same coding quality. Significant progress. For example, a typical application of H.264/AVC is a compression ratio ((10) plus ^) that requires a relatively souther wireless image, such as for use in an image mobile phone. Deblocking filters are commonly used with block-based digital image compression systems. The deblocking filter can be applied to a compression loop, where the filter is applied to the encoder and

On I290H. In another case, the block m can also be compressed in the loop; ίϊ = (4). A typical deblocking filter is in the low-pass filter ^ 乍, during which the block conversion coding (such as the dct) patch 2 block filter can be reduced in the decompressed image. The negative visual impact of a ΐ ΐ ( Μ — '), the value is ^: requires the image encoding H and / or the decoder to perform a large number of complex calculations. In order to make the resulting output image as close as possible to the original input image, ^ = worry = to reduce the block effect by deblocking the filter. Since the XCT is used to perform the DCT and quantization steps for the margin (_code, and therefore the block effect is generally better than H.264/AV, this prior standard can generally be chosen to use == 2, 264 saz In the implementation of dct and quantification to use = : prime early 兀 'Ting will produce more block effect:: VC recommended C峨, effective de-zone ::: [Summary] Upper and lower device and method The prior art has solved the problem of η, and the disadvantages. The pipeline type deblocking filter is more than the buffer, and the filter-wave engine ^ Qing, the 官 'the official line control unit' and the filtering engine are paired. After the block conversion processing, the material block performs filtering with the I29〇mpifdoc=small&block effect implementation method including: performing the (four) wave for this block, and after performing the first-edge filtering, the pair does not exceed ϋ The third block performs a third edge filtering, where the third edge is perpendicular to the edge of the strip

In order to make the present invention and other objects, and the advantages and advantages of the present invention are as follows: </ RTI> </ RTI> </ RTI> <RTIgt; The AVc off-process deblocking waver, which includes the high-speed mobile application 4, the quantity === 彳=execution before the occurrence, can also be referred to the processing of the current picture and the method of the method, which The reference number 10π is not shown. The encoder 100 includes an image wheeling end 112, the forward input of the bean fff 114 for performing signal communication. Addition block == zone overall conversion to provide the coefficient (ent-y-coding) block 118, with: ! horse, to "entropy coding _ with the binding of the smoke code to provide the rim / block 116 and transfer to - Block W in the loop! 20 and inverse transform (4) Trans.) Block 122. This 1290 instrument if.doc and inverse transform block 122 are coupled to an add block 124, and the add block 124 is recoupled. An intra-frame prediction block 126 is received. The intra-frame prediction block 126 is switchably coupled to a switch 127, and the switch 127 is further coupled to the second input of the addition block 124. At the same time, it is also lightly coupled to the inverting input of the add block 114. The output of the add block 124 is coupled to the limit deblocking filter 140. The restricted deblocking filter 140 is coupled to the frame store 128. The storage &amp; storage 128 is coupled to the mobile compensation (moti〇nc〇mpensati〇n) block 13〇, and the motion compensation block 130 is coupled to the second selected input image input 112 of the switch 127 and coupled to the motion estimation ( M〇ti〇nestimati〇n) block 119 to provide a motion vector (moti〇n vect〇r). The restricted deblocking filter 140 is coupled to the mobile estimation The second input of the block 119. The output of the motion estimation block 119 is coupled to the motion compensation block 13A and is coupled to the second input of the entropy coding block 118. The image input 112 is coupled to An encoding control block 160. The encoder control block 16 is coupled to the control inputs of the zones 116, 118, 119, 122, 126, 130, and 140 for providing control signals to control the encoder 10 ( Operation A. Referring to Figure 2, a representation of a decoding boundary with a de-blocking chopper in a loop is generally indicated by reference numeral 200. This decoder 2 includes an entropy decoding (entropy-decoding). The ghost 210 is configured to receive an input bit stream. The entropy decoding block 210 is coupled to the scaling and inverse conversion block 222 in the in-loop component 22 to provide coefficients. &amp; scaling and inverse conversion block 222 Light ^ = normal block 224, addition block 224 and then to the intra-frame prediction block. The intra-frame prediction block 226 is switchably coupled to the switch 227, the switch I29〇mPi, d〇c = == The second input of the addition block 224 is simultaneously consumed by the addition limit area ^= input. The output of the addition block 224 is to the σσ ghost The filtering source 240 is used to provide an output image. The stored block filter is lightly connected to the frame storage 228. The frame coupling: 2 2 2 f is connected to the motion compensation block 230, and the motion compensation block 230 is selected: A second selection input to 227. Entropy decoding block 2 compensates for the second input of the block to provide a motion vector. The smoke is then secreted to the decoder control set 262 to provide control information. The decoder control block 262 is coupled to the blocks 222, 226, 23, with the control input of ==, the wire transmits the control signal and controls the decoder. The area is broken. Figure 3 'With post-processing (P〇 St-pr〇CeSsing) goes to the second embodiment of the decoding &amp; Said. The coder 300 includes an entropy decoding block 310' for receiving the input bit 々丨L. The quotient decoding block 310 is coupled to the in-loop component = block 322. The scaling and inverse conversion block (4) =: The internal force port block 324 is again connected to the intra-frame prediction block Μ6. The frame Z block 326 can be switched _ to the switch 327, and the switch milk is again engaged. The second input end of the ί block 324 is simultaneously lightly connected to the reverse wheel of the addition block 3H. The output of the 2 block 324 is lightly connected to the limit deblocking wave 340 = for outputting an image. The addition block 324 is again _ to the frame storage. W 存 H 328 is reduced to the mobile compensation block, and the mobile compensation 33 〇 is lightly connected to the second selection wheel of the switch 327. The smoke decoding ι〇 I29043S8pif.doc i compensates for the second input of the block 330 to provide the mobile direction code block 31 and is coupled to the decoder control block 362 control signal. The decoder control block 362 _ to the blocks 322, 326, the offset input terminal 'is used to transmit the control signal and controls the decoder every = 4 = display, an encoder with a de-blocking filter in the loop == = It is indicated by reference numeral 400. The encoder 4 includes an input port 412' for receiving an input image image having a plurality of macroblock forces = rk). Image input 412 is coupled to the two = wide (four) input to perform signal communication. The addition block 414 block 416' is used to receive the margin, perform the discrete cosine transfer, and receive the quantization (Q) for the subtraction. Function block 416 trr block 418 is used to perform entropy or variable length coding (v: blelength(7)ding ' VLC) to provide an output bit stream. Block 416 is again reduced to inverse quantization 0, and iL: idct block 422. Dequantization and Detachment Receive: =ίΐί to block filter. The deblocking filter 440 is coupled to the material 428 to provide an output image image. The frame memory 428 == the first input of the module 429 to provide a reference frame to the predictive module 429 includes a motion compensation block 43 〇 . Block 419: ! ' Please. The frame storage 428 is in turn coupled to the _ input terminal of the mobile estimate = 9 to provide a reference frame to the block. Like the round ir and 412 is transferred to the second loss of the motion estimation block 419

10 12904 Secret 8pif.doc Into the end to provide a mobile vector. The output of the second wheeling block 419 of the prediction module 429 is coupled to the center I' / of the motion estimation block 419 and coupled to the motion compensation block 430. Two inputs. The inverse wheel number of the = area of the frame 2 entropy coding block 418 =

In the operation of the encoder gamma shown in FIG. 4, for example, an input map, or 峨 is divided into a plurality of macroblocks (_. Then, each: macroblocks are (10) 16 pixels' and each macroblock is in order Input into the ^, vc system. The prediction module 429 looks at all the macroblocks of the reference frame (which is the frame of the first wave over the chopping wave), and outputs the macro block output similar to the input Nat attack as a prediction. Therefore, the pixel value of the prediction signal is closest to the current MB. The margin is the difference between the pixel value between the current MB and the prediction signal. The coefficient is generated by performing DCT and quantization operations on the margin. In comparison, the data size of the coefficients is greatly reduced.

The coefficients may be encoded into an output bit stream by entropy encoding (as in block 418). This output bit stream can be stored or transferred to other systems. In addition, the coefficients can be converted to margins by IQ and DCT operations. The margin is added to the prediction signal' and converted to recon data. Recon_data has a block effect or block effect, which is generated by the boundary of a macroblock (16x16 pixels) or a block (4x4 pixels). Referring to Figure 5, a filtering sequence based on the H.264/AVC standard is indicated by reference numeral 500. The filtering order 50 〇 includes horizontal filtering of vertical edges I29 〇 mpif.d 510 and vertical filtering of horizontal edges 520. The H.264/AVC standard requires that all macroblocks of an image be filtered. The filtering is performed in columns of macroblocks (MB), behavioral basis, 4x16 pixels and 16x4 pixels, respectively, wherein the macroblock is 16χ16 pixels, and each block is 4&gt;&lt;4 pixels. The deblocking wave order based on the Η.264 standard is as follows. For luminance, as shown at 510, filtering is performed on the four vertical edges from the left edge. This is called horizontal filtering, wave. As shown at 520, filtering is performed on the four horizontal edges in the same manner from the upper edge, which is called vertical filtering. Chromanation filtering also follows the same ordering. Therefore, two vertical edge 510 filtering and two horizontal edge 520 filtering are performed on Cb and Cr, respectively. Deblocking is often a sloppy process due to frequent memory accesses. To perform filtering on the vertical edge 2, the left (previous) and right (current) column data accesses are performed on the buffer. Therefore, 4x16 pixel data access is applied twice for each edge. According to the • H.264/AVC standard, horizontal filtering (luminance steps 1, 2, 3, and 4) is completed after 'vertical filtering (luminance steps 5, 6, 7, and 8). To perform vertical filtering, the data previously accessed in the horizontal filtering step must be applied. All 4 χ 4 pixel blocks in the 16x16 pixel macroblock are stored. Therefore, the wave logic size and filtering time increase. For the current embodiment, the deblocking filtering time of a macroblock should be limited to 500 clock cycles to view the high/month clear image. To achieve this ratio, the luminance (lu·) wave and the chrominance () 12 I29〇4MPi, d〇c lines are used to complete the filtering in time. Unfortunately, the side-by-side execution of the road, so the focus of the filter and the chromaticity of the surface of the electricity: to the invention of the 2 out of a kind of pipeline filter sorting, degree or yellow stitching 61 〇, order _ including bright. ^ &amp; Chromaticity Filter Ordering 620, and Red Symplectic == 63 (1° Brightness, Wave Ordering 61G includes Luminance Filtering Step 1 μ 区 区 block p. Blue chrominance filtering order 620 includes

=ίΐίSteps 33 to 4, for blue chroma blocks Q to T

48 is used for ΓΓί sorting (10) including red chroma filtering, wave steps 41 to 48 for red chrominance blocks U to X. J -==Based in the middle: Area:; Consider:: _ Open block (if in terms of brightness is above: two thin rows or columns (such as 'pair = line. Apply the filter order currently described as saying 4X16 pixels, for chrominance, ke like symplectic): as several segments (eg, 4 segments for brightness, for color sound = segment). As specified in the H.264/AVC standard, this row "" 仗 left to right, top to bottom order. 寸, because the filtering operation is a block (tree pixel, a l L body access times reduced. In addition, because = use the data between adjacent blocks Correlation, and in the 2nd 13 129043 &amp;p if.doc factory poison 600 r, in a block (4x4 pixels) = two = times to perform filtering. For example, the area is changed to two (four),) /, Filtering is performed in this order 12 and 13. In addition, Λ F 20 is for &amp; block 1 and §, edge 21 is the upper edge). The edge filtering process of block F is as follows: The upper edge of block E:: prime = updated to the left side register for storing: medium. Second, the pixel value of block G is sent to the engine to perform filtering using the area buffer. Third, the filter county is executed by the reference + f # right edge 12. Block F_ == side register, and block. The new pixel value is more in the sub. Fourth, the pixel value of block B is buffered from the top to perform the test. Block B _ is updated to medium. Sixth, the new pixel value of the &quot;V magpie block F is updated to the left register wave. H21 will perform the pixel value of the filter S in the edge filtering process of the block without having to store it in the memory or calculate the new pixel value from the memory to the memory Ji!1, without having to store it. Since the U5 memory is retrieved from the memory, the temporary memory access can be updated immediately and the (4) block-based small 14 I29043 (88pif.d〇c filter, wave unit is used, so the filter logic is simplified. And the filtering time is shortened. It should be understood that the filtering ordering for luminance, red chrominance and blue chrominance is separately defined. • That is, luminance filtering can be performed before, after or between red and blue chrominance filtering, Red chrominance filtering can be performed before or after blue chrominance filtering, luminance filtering, or both. Therefore, in addition to the exemplary 4:1:1 Υ/Cb/Cr format, the currently described block filtering ordering It can also be applied to other types of block formats. As shown in Fig. 7, a deblocking block filter according to an embodiment of the present invention is denoted by reference numeral 700. This deblocking filter 7〇〇 A buffer or current memory 710 is included for storing the reconstructed material of the current macro block (mb). The buffer 710 and the filtering unit 712 perform two-way communication to provide current data and MB start signals to the filtering unit. Filter The element 712 includes an engine 714, a register block 7^, and a finite state machine (FSM) 718. The FSM 718 in the annihilator 70 712 is connected to the current data controller 72A for signal communication. To provide the FSM status signal and the counting signal to the current data controller ^ πο. The current data controller 720 is then connected to the current memory 71 〇 = line 2 number k to provide a memory signal or static random access memory (He He random The access memory (SRAM) control signal is given to the memory. When the reconstructed data (which is the prediction signal plus the margin) is stored in the current record, the filtering is performed. The filtering unit 712 and the filtering boundary strength (fiitering) The B〇undary Strength, BS) generator 722 is connected to perform signal communication to provide status signals, count signals, and flag signals to the state generator.

15 129043?&amp;8pif.doc The ancestor 722 is then connected to the Quantizati〇n - Parameter of neighbor block (QP) memory 724 for communication. The filtered boundary strength generator 722 is in turn coupled to the filtering unit 712 for signal communication to provide parameters to the filtering unit. Filtering unit 712, in turn, is coupled to proximity control 726 for signal communication to provide a sad alarm and count value from FSM 718 to the proximity controller. Control crying 72 disks Near memory or buffer 728 is connected for signal communication (10) storage adjacent = block. The proximity buffer 728 receives the memory signal from the controller 726 or the sneak edge machine access memory (SRAM) control signal. The buffer 728 is coupled to the filtering unit 712 for signal communication, provides first neighboring data to the filtering unit 712, and receives second neighboring data from the filtering unit. The filtered boundary strength generator 722 is in turn coupled to the proximity controller 726, the top controller 730, and the direct memory access (this mem〇ry, DMA) controller 734 for signal communication to provide parameters to the controllers. The filtering unit 712 is further connected to the top controller 73A for signal communication to provide status signals and counting signals to the top controller, and is connected to the DMA controller 734 for signal communication to provide status signals, counting signals and chrominances "δ". The top controller 730 is connected to the top memory 732 for signal communication to provide an SRAM control signal to the top memory. The top memory 732 is coupled to the filtering unit 712 for signal communication to provide the first - top data and receiving second top data from the wave unit, wherein the top data is for vertical filtering. The DMA controller 734 is coupled to the DMA memory 736 for signal communication to provide SRAM control signals to the DMA memory. The unit 712 is also connected to the 129043 8pif.doc DMA memory 736 for signal communication to provide the filtered data to the DMA memory. The top memory 732 and the DMA memory 736 are both connected to the switching unit 738 for signal communication, and the switching unit 738 is further connected to the DMA bus interface 740 for signal communication to provide filtered data to the DMA bus. The filtered data is transmitted to the DMA through the DMA bus interface 740. The illuminating image of the aging &amp;

It is indicated by reference numeral 800. This pipelined architecture can be used in conjunction with efficient wave sequencing to further reduce inspection time. This deblocking filter is formed into a layer 4x4 block segment 801 and a 4x1 pixel segment 8〇2. The 4x4 block pipeline segment 801 responds to fsm 718 in FIG. The pipelined architecture includes a first-block pre-fetching and finding step (10) by which the neighboring lean material is pre-fetched from the neighboring buffer 728 of FIG. 7 into the temporary buffer 710 to read the current data, and by generating Pixel value brother looking for BS declining moxibustion. Potential _ ^ ^ 盥筮Carpel> Number of first block filtering and storage steps 812

= - Block pre-fetching and finding step 81 () section Filtering and storing step 812 pairs (4) = stored in the buffer memory. After executing 1 新£新' and taking the result 810, the material is pre-extracted and the step is searched for the remaining blocks to perform the step 814, and after the step 812, the execution is performed. After the first block-block filtering and post-storage, the remaining blocks are executed 818 and so on. ::: Waves and storage steps 816 'Rare step 814 and block-block filter "Second two blocks pre-extract and search and The storage step 816 is overlapped. The sub-step 812, the second block filtering I29〇4S^pif.d〇c 4xl pixel edge pipeline segment 802 is in response to the pixel edge pipeline segment 8〇2 in FIG. 7 including: the first engine 4 In this case, the first 4×1 searching step 822 for pre-extracting the pixels, after performing step 820, finds X, :beta and tcO turns to the first column of the first block to save steps. 824 'The first 4x! column used to filter the second = ^ a 4&lt;4 block after performing step 822. The pixel edge pipeline segment 'children a 4x1 pixel pre-extraction step 83 匕 · 叫 寻找 寻找 寻找 寻找With step coffee power most 822 father Feng; second and storage step 834, in step 832 彳 ^ brother 2 4X1 filtering also includes Flute II / , your main i is more stubborn. In addition, the pixel segment 802 繁 - 4 ! ί pixel extraction step 840, with step 832 A · 弟二 4x1 brother find step 842, with step six Tian ·, and, and The storing step 844 is performed after the step=, and the second 4x1 m^2. There is also a fourth 4x1 pixel pre-k taking step 85〇, and step 84 step 852, overlapping with step 844; and, x brother finds 854, after step 852. The fourth four 4x1 filtering and storing step 1 pixel # 802 pre-extracting step ^ finding step 822 and pre-extracting step 83 〇 = J finder second pre-extracting step 814 execution period; in the 4X4 area Block 824, search step 832, and pre-filtering and storing step 822 and pre-fetching (four) please 840 after searching for the second filtering step (4) during execution == in block number 4 operation 'because The pre-i step of the 4X1 pixel segment, the steps of the parameter, and the filtering and storing steps are all performed in the 12904 pifdoe case of the block segment, and the execution of the chroma is performed after the demand filtering. Filtering. Therefore, only = i) f Wei, the green domain is correcting the hardware Inch. Press the parameter t Kui, to 'r L 5 · seeking mainly to illustrate the present illustration

^ λ J knows that the new pixel value of the temporary state is updated to the current (the second; the memory ^, and the new pixel value of the adjacent register is updated to the adjacent temporary 4, 6 execution filter will be executed after the water gradient Wei (such as edge ^ 12, etc.) counts different results. For example, for a circled edge number 4, the new pixel value of the current register (ie, the block Β) is further 2 adjacent to the temporary (four) towel. (4), the block C pixel value is directly loaded from the current memory ^. In the edge 4 filter, the wave (which happens to follow the edge 3 filter) two 'near the temporary fiscal block block Α pixel value. Thus, in the same way Counting 8 of the 32 edges (ie, edges 4, 6, &amp; 14, 2, 22, 28, and 30). Referring now to Figure 9, a filter circuit is referenced to 9 The filter circuit 900 includes a finite state machine (FSM) 91A that is connected to the engine 912 for signal communication. The FSM 910 receives the MB start signal (MB_start) and provides a chroma flag signal (Chr〇ma Flag), fsm count signal (inFSM-cnt), line count signal (Hne-Cnt), and ?81^ status signal (1^]\4^&am p;16).8]\4 is further connected to an input switch or a control input of a multiplexer 914 for signal transmission 19^9〇4^pifdoc k, and the multiplexer 914 receives the first adjacent data ( Neig-datal), first top data (top-datal), or current data (current-data), and each time one of these kinds of data is provided to the temporary register 916. The register 916 and the output switch The 918 is connected to perform signal communication to provide second neighboring data (neig-data2), second top data (top-data2), or filtered data. The engine 912 has: an input terminal for receiving the BS and the parameter. a signal for receiving a current neighboring pixel and a current pixel (p and q) input from the register 916; and an output for providing the updated neighboring pixel and pixel (p, and q,) The output signal is sent to the register 916. In this specification, the MB-START and Na-end cutters are marks indicating the start and end of the 1MB filter, and the output apostrophe of the fsm has MB-END. Chroma-Flag indicates the brightness. Or the color of the mark. FSM a state is the output of the FSM , indicates the horizontal position of the current 4x4 block in 16χ16ΜΒ. The inFSM-cnt signal indicates whether the hi pixel pipeline segment in the block completes the operation. The line-ent signal indicates the vertical position of a block in one MB. Neig-datal is a 4xl pixel neighboring data used for rain MB level crossings. Tonne-Yang] is the (four) pixel data stored in the -buffer for the next MB horizontal filtering. Top-datal is the top information for the vertical filtering of the current block. t叩一-2 is stored in a buffer for straight two 1 - ta is the current 4x1 pixel data. Data is the pixel data of the filtering operation after the filtering operation, respectively, the adjacent 4 pixels after filtering and after chopping, and the current co-pixels before and after filtering. The scratchpad includes - temporary it 20 1290 Confucian Pif.d0c column. The engine performs filtering operations based on the state of the FSM. As shown in Fig. 10, a filter circuit with other blocks is indicated by reference numeral 1000. This filter circuit 1000 includes an engine 1 〇 12 for receiving the current neighboring pixel (p) from the multiplexer (MUX) 1010 and the current pixel (q) from the MUX 1011. The engine 1〇12 is connected to the mux 1013 and the MUX 1014 for signal communication. Μυχ 1〇π is connected to the 4x4 block register array 2 1016 for signal communication. The 4x4 block register array 2 1016 is connected to the MUX 1018 for signal communication. The MUX 1018 provides neighboring data (neig_data2) to the neighboring memory (NEIG-MEM) 1020, and the neighboring memory 1〇2〇 provides another neighboring data (neig_datal) to the MUX 1010. The 4x4 block register array 2 1016 is further connected to the top memory (TOP-MEM) 1022 for signal communication, and the top memory 1022 is connected to the MUX 1024 for signal communication. The MUX 1024 is then coupled to the 4x4 block register array 1 1026 for signal communication. The 4x4 block register array 1 1〇26 is connected to the MUX 1028 for signal communication, and the MUX 1028 is connected to the bus interface (BUS_IF) 1030 for signal communication to provide filtered information to the interface, wherein the interface and the DMA are provided. The memory is connected for signal communication to provide a deblocking output signal (DEBLOCK_OUT).

The filter circuit 1000 further includes a pair of current memories (CURRJV1EM) 1032 for receiving the reconstructed data (RECON_DATA). Each current memory 1032 is connected to the MUX 1034 for signal communication, and the MUX 1034 is connected to the MUX 1011 for signal communication to provide current data (curr_data) to the MUX 1011. The current memory 1032 is further connected to the FSM 21 l29〇mP, doc 1036 for signal communication to provide an activation signal (mb_start) to the FSM4X4 block pipeline architecture 1036. The 4SM 1036 is connected to the controller 1038 for signal communication to transmit the signal FSM__state, Line_count and ChiOma-flag are provided to the controller 1038 and receive the signal inFSM_count from the controller 1038 to the 4x1 pixel pipeline. Controllers 〇38 are coupled to the control inputs of each of the MUXs 1010, 1011, 1014, 1018, 1024, 1028, and 1034 for signal communication to control the MUXs in response to FSM-state, line-count, Chroma-flag, and inFSM-count signal. In operation, the MB_START signal is generated when the recon_data signal is stored in CURR_MEM and the filtering operation begins. The FSM receives the control signal inFSM_cnt from the 4x1 pipeline controller to see if this 4xl pixel pipeline segment has completed its operation. Since the same filtering and wave engine are used for luminance filtering and chroma filtering, the Chroma_Flag signal is required. The engine filtered data is transferred to the memory or DMA by BUS_IF. Referring to Figure 11, a timing diagram of a pipelined architecture is indicated by reference numeral 1100. The timing diagram 1100 is shown as signals HCLK, MB_start, line_cnt, FSM, inFSM_cnt, Filtering-ON, BS, ALPHA/BETA/TCO, p, q, filterSampleFlag, filtered_p, and filtered_q, respectively. Relative timing. The timing diagram 1100 is further illustrated as a 4x4 block pipeline segment, which includes: Step 1110, pre-fetching and searching for the BS to the first block; Step 1112, performing filtering and storing the filtering result to the first block; Step 1114, searching for an alpha , beta and tcO parameters are given to the first block, wherein step 1114 and step 22^29〇43 collapse pif.doc • H〇, 1112 overlap; step 1120, pre-fetch and find BS to second • $; step 1122 Performing filtering and storing the filtering result to the second block; 2^1124' searching for the aipha, beta, and tc〇 parameters to the second block, where 1124 overlaps steps 1120 and 1122; step 113〇, pre-fetching $evaluation The BS is given to the third block; in step lm, filtering is performed and the filter is stored, and the second block is sent to the second block. In step 1134, the alpha, beta and tcO parameters to the third block are searched, wherein step U34 overlaps steps 1130 and 1132. • In addition, step Π20 for the second block is used with the step 1122 for the first block and 1114 for the parent, the step 1124 for the second block overlaps with the step 1112 for the first block, In step 113 of the third block, the second block is overlapped with step 22. Referring now to Figure 12, a filtering method proposed in accordance with the filtering block ordering of the present month is generally indicated by reference numeral 12?. A macroblock is grouped into a luminance portion 12〇2, a first chrominance portion 1204, and a second chrominance portion 12〇6, each portion having a vertical edge, starting from the left edge m=0, and each portion With a horizontal edge, the upper edge starts at π=0. This filtering method 1200 includes activating block 1210 that is initially set to Chr〇ma=No, m=0, and η=〇. Boot block 1210 passes the control signal to functional block 1212, which performs filtering on the edge of the vertical 4χ4 block of the MB of m=〇. The function block 1212 passes the control signal to the function block U14, which performs filtering on the edge of the vertical 4x4 block of the MB of m=l. The function block 1214 passes the control signal to the functional block 1216. The function block 1216 performs filtering on the horizontal 4x4 block edge of the MB of m = 0 and passes the control signal to the decision point 1217. 23 1290 Confucian pif.doc Decision point 1217 determines if the block is a chroma block, and if so, • passes the control signal to function block 1218. If the block is not a chroma block, then the control signal is passed to the function block 122. Function block 1220 performs filtering on the vertical 4x4 block edge of the MB of m = 2 and passes the control signal to functional block 1218. The function block 1218 performs filtering on the first horizontal edge of mb of m = 1 and passes the control signal to the decision point 1222. Decision point 1222 determines if the block is a chroma block, and if so, # passes the control signal to decision point 1224. Decision point 1224 determines if the block is an end block in the MB. If so, the control signal is passed to end block 1226. If not, decision point 1224 passes the control signal to decision point 1225. Decision point 1225 determines if n = 1. If η=ι, reset it to n=0. If η is not equal to 1, η is incremented by 1. After decision point 1225, the control signal is passed to functional block 1212. On the other hand, if decision point 1222 determines that the current block is not a chroma block, then the control signal is passed to power block 228. The function block I228 performs filtering on the vertical 4&gt;&lt;4 block edge of MB of m=3, and passes the control signal to the function block 123A. The function block 1230 performs filtering on the third horizontal edge of the MB of m = 2 and passes the control signal to the functional block 1232. The function block 1232 then performs filtering on the fourth horizontal edge of the MB of m = 3 and passes the control signal to the decision point 1234. The judgment point 1234 determines whether n=3. If n=3, reset it to n=〇 ' and set chromates. If η is not equal to 3, then η is incremented by 1 24 I29〇43^if.d〇c. Judge 1J; Ho &gt;· 1212. According to this description, the == signal is transmitted to the functional block. The above and the other ones of the invention can easily find the extensible Ah II and the advantages. For example, it should be understood that the present invention shortens the filtering to: the embodiment of luminance filtering and chrominance filtering, before the step-by-step, the heart luminance filtering can be filtered in red, blue chrominance, and the bright red chrominance filtering can be in blue color. Filtering is filtered again, or both before or after. apart from

Applied to other types of block formats. In this specification, the preferred edge chopping order of the giant tilt block based on the =264/A^C standard is disclosed, but it should be understood that the 'per-block general filtering ordering (wearing the implementation of the St flat edge) can be applied. Various Other Types It should be understood that the present invention can be implemented in various forms of hardware, software, firmware, special purpose processors, or combinations thereof. Moreover, the software preferably operates as an application that is substantially present in the program storage device. An application can be uploaded to and executed by a machine containing any suitable architecture. Preferably, the machine can be hardware-equipped (such as one or more central processing units (CPUs), a random access memory (RAM), and input/output (input/output, I/O) interface] performs operations on the computer platform. The computer platform can also include an operating system and microinstruction code. The various methods and functions described in this specification can be either part of the microinstruction code or an application of 25 1290 (5). c f, any combination of them - part of 'they can be executed by the cpu external data storage list = page him, set:: connected to the computer platform, such as the actual part of the amount or processing function block, /,, the connection according to the implementation The design method varies from case to case. Qualifications: 2: The implementation of the whistling embodiment is as described above, but it is not intended to be familiar with the artist, without departing from the spirit of the invention.

The latter is considered; the protection of the present invention [Simple description of the drawing] In the following table, the pipeline deblocking filter and the square 1 are proposed according to the following embodiment, and the material phase_component can be represented by a reference symbol. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 is a schematic block diagram of an embodiment of a coded cry with an in-loop deblocking filter. Port Figure 2 is a schematic block diagram of an embodiment of a decoder with an in-loop deblocking filter and an encoder not shown in Figure 1. 3 is a schematic block diagram of an embodiment of a decoder with a post-processing deblocking filter. 4 is a schematic block diagram of a codec with an in-loop deblocking filter, which conforms to the H.264/AVC standard. Figure 5 is a schematic diagram showing the basic filtering sequence based on the H.264/AVC standard. 6 is a schematic diagram of a filtering sequence that satisfies the H.264/AVC standard requirements and is proposed in accordance with an embodiment of the present invention. FIG. 7 is a schematic block diagram of a deblocking filter according to an embodiment of the invention. (d) Figure! / = is an unintentional timing diagram of a pipelined architecture proposed in accordance with the present invention. A f 9, , g is not a schematic block diagram of a filter circuit according to an embodiment of the present invention. FIG. 10 is a schematic block diagram of a filter and related blocks according to an embodiment of the invention.

Figure Η % is not a partial schematic timing diagram of a pipelined architecture block proposed in accordance with an embodiment of the present invention. Figure 12 is a schematic flow diagram of a sorting filtering method proposed in accordance with the present invention. [Description of main component symbols] 100, 400 · Encoders 112, 412 · Image input terminals 114, 124, 214, 224, 314, 324, 414, 424 · Addition blocks 116, 416: Function block 118, 418: Entropy coding blocks 119, 419: motion estimation blocks 120, 220, 320 • In-loop components 122, 222, 322: inverse conversion blocks 126, 226, 326, 426: intra-frame prediction block 127, 227, 327 · · switch 128, 228, 328, 428 · frame memory 27 I29 〇 4si 7%) if.doc 130, 230, 330, 430: motion compensation block 140, 240, 340, 440, 700: Deblocking Filter 160 - Encoder Control Blocks 200, 300: Decoders 210, 310: Entropy Decoding Blocks 262, 362: Decoder Control Block 422: Inverse Quantization and Inverse Discrete Cosine Transform Block 429: Prediction Mode Group recon_data: Reconstruction data 500, 600: Filtering sequence 510: Vertical edge 520: Horizontal edge 1, 2, 3, 4, 5, 6, 7, 8: Filtering step 610: Brightness or yellow filtering ordering 620: Blue chrominance filtering Sort 630: Red Chroma Filter Sort 1 to 32: Luma Filter Steps 33 to 40: Blue Chroma Filter Steps 41 to 48: Red Chroma filtering steps A to P: luminance blocks Q to T: blue chrominance blocks U to X: red chrominance block 712 ··filtering units 714, 912, 1012: engine 28 12904 temperature pif.doc 716: temporary Storing blocks 718, 910, 1036: finite state machine (FSM) 720 (CURR_CTRL): current data controller 722 (BS__GEN): filtering boundary strength generator 724 (QP_MEM): neighboring block quantization parameter memory 726 (NEIG) - CTRL), 730 (TOP_CTRL), 1038: Controller 710 (CURR-MEMX2), 728 (NEIG-MEM), 732 (TOP-MEM), 1020 (NEIG-MEM), 1022 (TOP-MEM), 1032 ( CURR_MEM): memory 734 (DMA_CTRL): direct memory access controller 736 (DMA_MEMx2): direct memory access memory 738 · · switching unit 740 (DMAJBUS_IF): DMA bus interface SRAM CONTROL, FSM-STATE, INFSM—CNT, NEIG—DATA1, NEIG—DATA2, TOPDATA1, TOP_DATA2, PARAMETER, FSM, LINE—CNT, CHROMA—FLAG, FILTERED DATA, OURR-DATA, MB—START, CURRENT-DATA, BS, p, p', q, q', ALPHA, BETA, TCO, HCLK, FILTERING-ON, FILTERSAMPLEFLAG, FILTERED_p, FILTERED-q: Signal 800: Pipeline deblocking filter architecture 8 01 · 4 χ 4 block segment 802: 4x1 pixel segment 29 1290 burst f.doc 810, 812, 814, 815, 816, 818, 820, 822, 824, , 830, 832, 834, 840, 842, 844, 850, 852, 854: steps 900, 1000: filter circuits 914, 1010, ion, 1013, 1014, 1018, 1024, 1028, 1034: multiplexer (MUX) 916 : register 918: output switch 1016, 1026: 4x4 block register array • 1030 (BUS_IF): bus interface 1100: pipeline architecture timing diagrams 1110, 1112, 1114, 1120, 1122, 1124, 1130 1, 1122, 1134: Step 1200: Filtering Method 1202: Luminance Section 1204, 1206: Chroma Section 1210: Boot Blocks $1212, 1214, 1216, 1218, 1220, 1228, 1230, 1232: Function Blocks 1217, 1222 1224, 1225, 1234: decision point 1226: end block 30

Claims (1)

Japanese repair (seven ^ replacement page revision date July 4, 1996 1290438 for the 94141084 Chinese patent scope 18678pifl.doc X. Patent scope: _ 1. A filtering method, using block conversion to the pixel poor area Wei Filtering to reduce blockiness, the filtering method includes performing a first edge chopping on the block and performing third edge filtering on the block of no more than three edges after performing the first edge filtering, wherein the third edge The silkworm is straight to the first edge. 2. The filtering method of claim 1, wherein the first edge is the left edge of the block and the third edge is the upper edge of the block. The filtering method of claim 1 further comprises performing a second edge filtering on the block not exceeding the edge of the rain strip after performing the first edge filtering, wherein the second edge is parallel to the first edge. The filtering method of claim 3, wherein the second k-edge is a right edge of the block. 5. The filtering method of claim 1, wherein the block comprises 4×4 pixel data. Shen The filtering method of claim 1, wherein the block is one of 16 blocks including a macroblock. 7. The filtering method as described in claim 6 of the patent application The blocks in the cluster block are sequentially filtered from left to right and from the top row to the bottom row. The filtering method described in claim 1, wherein the pixel data block includes multiple rows and more Column or a plurality of pixel vectors, the filtering method fork includes: pre-extracting adjacent block pixel data to the first register array; 1290438 18678pifl.doc pre-fetching current block pixel data to the second temporary memory to find the boundary of the current edge Intensity, back to the second ''乂 and the data and the pre-extracted current pixel data...., # near pixel please filter the method described in item 8 of the patent scope, including: advance, upper block pixel data to the third temporary storage The array of arrays includes the filtering method described in item 8 of the range, including: filtering the engine: pre-extracting the neighboring vector of the pixel (4) to the pre-extracted pixel data in the register array current The vector vector skin amount parameter gives the adjacent wave corresponding to the boundary strength of the current series; the neighboring vector (10) of the neighboring shirt (10) is used to perform the neighboring vector to the η. The filtering method in the register array further includes: feeding the filtering engine; and extracting the neighboring vector of the pixel data to the second and second register arrays to find the filtering parameter to the current and current vectors; °°4 The boundary strength corresponding to the neighboring vector corresponding to the filtering parameter and the neighboring current vector perform filtering 32 1290438 18678pifl.doc wave; storing the neighboring vector after the wave into the memory; and updating the filtered current vector to the first In the second register array. 12. If the scope of the patent application is 1st, the second temporary storage of the crying mausoleum, the squatting place. 'Thinking; the method' includes: the column performs the update; u (four) (four) touches the first-stage register, saves the update, The first register array is in the memory; and the first register array is stored in the memory, and the pixel-block is pre-followed to the second register array. 13=Application of the filtering method described in the 1Gth scope of the patent, further comprising: when the filter is used to obtain the first neighbor vector, the second pixel data neighboring vector is pre-extracted from the first register column to the filtering engine. When searching for the filter parameter to the first current vector, 7 pre-extracting the current vector of the second pixel data from the second register array to the filtering engine; when performing filtering and wave on the first neighbor vector and the first current vector, searching The wave parameter gives a second neighboring vector corresponding to the boundary strength of the current block and a second current vector; ^ performs filtering on the second neighboring vector and the second current vector corresponding to the filtering parameter; μ updates the second filtered neighboring vector Going to the first register array; and updating the second filtered current vector to the second register array. 14. The filtering method of claim 12, further comprising processing a block pipeline of the second pixel data block. 15. The filtering method described in item n of the patent application, block tube 33 1290438 18678pifl.doc The line processing method includes: pre-fetching the second block pixel data to the first register array; and finding the boundary strength of the block. 16. The method according to claim 15, wherein the block pipeline processing method further comprises: pre-fetching a second pixel vector from the block when searching for the filter parameter to the first pixel vector; When at least one of the two operations of filtering and storing the first pixel vector is performed by one pixel vector, the filter parameter is sought for the second pixel vector. 17. The filtering method of claim 15, wherein the vector pipeline filtering further comprises: pre-fetching another pixel vector from the block when looking for the filtering parameter to the previous pixel vector; and in the previous pixel When the vector performs at least one of the two operations of filtering and storing the previous pixel vector, the filter parameter is sought for another pixel vector. 18. The filtering method of claim 1, wherein the pixel data block comprises a row, a column or a vector having a plurality of pixels, the filtering method further comprising a pixel pipeline that performs filtering on the plurality of pixels. 11 The filtering method according to claim 18, wherein the pixel pipeline filtering comprises: pre-fetching a first pixel from the pixels; and searching for a filtering parameter to the first pixel; 34 1290438 18678pifl.doc 月日修(冬) The positive replacement page performs filtering on the first pixel; stores the first pixel; when taking a pico to the first pixel, preliminarily performs at least one operation of performing the second pixel filtering and the first pixel from the pixels When the job search wave parameter is given to the second image.线 线 渡 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利 专利In the middle of an operation, look for the filter parameters to another _pixel. 21. An official line deblocking filter for performing filtering on a pixel data block at a block conversion to reduce blockiness, the pipelined deblocking filter comprising: a filtering engine; a register, the signal communication with the filtering engine; an official line control unit, performing signal communication with the filtering engine; and a finite state machine for signal communication with the pipeline control unit. U. The pipelined deblocking filtering described in claim 21 of the patent application is combined with an encoder for rasterizing the pixel data as a plurality of block conversions, wherein the pipelined deblocking filter is configured to give The block conversion of the reconstructed pixel data performs filtering in response to the block conversion coefficients. 23) The pipelined deblocking filter 35 1290438 18678pifl.doc described in claim 21, combined with a decoder for converting the coded block for reconstruction, the pipeline data is _ Filter &quot;Configuration=Rebuild the block conversion of the pixel data to perform filtering. _ ^ 24. As described in the application of Paragraph 2i, the finite state machine is configured to control the wave block pipeline segment at the block. Deblocking filter 25. The apparatus described in claim 21, wherein the engine is configured to control the wave segment of the &amp; block. The position of the line-to-zone filter is as follows: Tube Ϊ1丨 η η to control the pixel of the pipelined deblocking filter to perform filtering on the ^ &gt;is ~ A ', . data block, first filter the three edges of the zone, and then perform the wire-edge lag wave and then A block of no more than one edge performs a third edge filter, wave, and the third edge of the towel is perpendicular to the edge of the step. I computer I / one / kind of computer readable storage media, can be read by the computer, including light you: 亍Ob々 program to execute the program step to perform filtering on the block data block processed by the block, The program steps include: the block performs the first edge filtering; and after the edge filtering of the 亍, the area is not more than three edges. 1290438 18678pifl.doc The third edge of the wave is 'the third edge is perpendicular to the first - 28 · If you apply for the k-edge. Step 2: The storage medium described in item 27 of the profit area, the program performs a second edge filtering on the block not exceeding two edges after performing the first edge filtering and wave, wherein the second edge is parallel to the An edge. 29. The storage medium of claim 2, wherein the pixel data block comprises a plurality of rows and a plurality of (four) _ pixel vectors, the program step further comprising: , and the neighboring pixels that should be pre-fetched by the street. Pre-fetching the neighboring block pixel. Pre-fetching the current block pixel. Finding the boundary strength of the current edge' data and the pre-fetched current pixel data. 37