JPEG2000 method based on channel number truncation and bit plane scanning

Abstract

The application discloses a JPEG2000 algorithm based on channel number truncation and bit plane scanning, which comprises the following steps: inputting a picture to be compressed; performing wavelet transform on the input picture, and dividing the obtained matrix after the transform into four subbands of LL, LH, HL and HH; performing code word allocation on each subband after the wavelet transform; performing bit plane channel scanning encoding on the LL subband, and performing one-time bit plane scanning encoding on other subbands; based on the code word allocation, truncating the encoding results of the other subbands; performing MQ encoding and outputting a compressed code stream. The application can acquire code words in advance, reduce unnecessary encoding, shorten the encoding time, use the one-time bit plane scanning mode in the bit plane encoding, greatly shorten the encoding time, reduce the context code words and the code stream transmission capacity.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically a JPEG2000 method based on channel truncation and bit-plane scanning. Background Technology

[0002] In 1996, the Joint Picture Experts Group (JPEG) committee proposed JPEG2000, which was released as an international standard in December 2000. Due to its scalable bitrate and high-quality reconstructed images, it has been applied in many fields. However, in the JPEG2000 encoding system, the EBCOT part is relatively slow, especially in the bit-plane encoding section, which requires scanning and encoding in groups of four columns across three channels. This process generates a large amount of context and consumes significant storage space. Furthermore, if the compression ratio is too high, the recovered image after truncation may be suboptimal. Summary of the Invention

[0003] The purpose of this invention is to address the problems existing in the prior art by providing a JPEG2000 algorithm based on channel truncation and bit-plane scanning, which can reduce bit-plane coding context, shorten scanning time, and improve coding efficiency. At the same time, the combination of this scanning method with the bit-plane-based truncation method can significantly improve the compression effect.

[0004] The technical solution to achieve the purpose of this invention is: a JPEG2000 algorithm based on channel truncation and bit plane scanning, the algorithm comprising the following steps:

[0005] Step 1: Input the image to be compressed;

[0006] Step 2: Perform wavelet transform on the input image and divide the resulting matrix into four sub-bands: LL, LH, HL, and HH.

[0007] Step 3: Assign codewords to each subband after wavelet transform;

[0008] Step 4: Perform bit-plane channel scanning encoding on the LL subband, and perform one-time bit-plane scanning encoding on the other subbands;

[0009] Step 5: Truncate the encoding result from Step 4;

[0010] Step 6: Perform MQ encoding and output the compressed bitstream.

[0011] Furthermore, in step 2, the wavelet transform uses the 5 / 3 wavelet transform, and its forward transform formula is:

[0012]

[0013]

[0014] Its inverse transformation formula is:

[0015]

[0016]

[0017] In the formula, For the odd-numbered rows / columns of the input matrix, For the even-numbered rows / columns of the input matrix, For the next even row / column of the input matrix, For the odd-numbered rows / columns of the matrix obtained by wavelet transform, For an odd-numbered row / column of the resulting matrix, The even-numbered rows / columns of the resulting matrix;

[0018] The above formula is used to perform row and column transformations on the input image to complete the wavelet transform.

[0019] Furthermore, the specific formula for codeword allocation in step 3 is as follows:

[0020]

[0021] in, It is the number of bytes allocated to the i-th subband. It is the target codeword count, which is the length that should be compressed and truncated. It is the number of encoded channels in the i-th sub-band. These are the wavelet subband weighting coefficients of the i-th subband.

[0022] Furthermore, the aforementioned =3*p-1, where p is the number of bit planes.

[0023] Furthermore, step 4, which involves performing a one-time bit-plane scan encoding on the non-LL subband, specifically includes:

[0024] Step 4-1: Perform bit plane scanning until all coefficients in the code block have been scanned;

[0025] Step 4-2: Perform one-time bit-plane encoding on each scanned column:

[0026] Step 4-2-1: Determine whether the coefficients of the current codeword sample to be encoded and its neighboring samples are all significant. If they are not significant, then perform run-length encoding, i.e., RLC encoding; otherwise, proceed to the next step.

[0027] Step 4-2-2: Determine whether the coefficient of the current codeword sample is significant. If it is significant, perform amplitude refinement encoding, i.e., MR encoding. Then determine whether the current column has been scanned. If the scan is complete, proceed to step 4-2-4; otherwise, repeat step 4-2 to continue encoding. If it is not significant, perform zero encoding, i.e., ZC encoding, and then proceed to the next step.

[0028] Step 4-2-3: Determine if the current codeword sample to be encoded is 0. If not, perform saliency encoding, i.e., SC encoding. Otherwise, determine if the current column has been scanned. If the scan has been completed, execute step 4-2-4. Otherwise, repeat step 4-2 to continue encoding.

[0029] Step 4-2-4: Determine whether the bit plane scanning is complete. If the scanning is complete, end the bit plane encoding; otherwise, start column scanning from step 4-1.

[0030] Furthermore, the specific method for significance determination in step 4-2-1 is as follows: if there is a bit '1' in the bit plane before the sample where the current encoded bit is located, then the significance of the sample where the encoded bit is located is '1', indicating significance; otherwise, it is '0', indicating insignificance.

[0031] Furthermore, step 5, which involves truncating the encoding result from step 4, specifically includes:

[0032] Compare the codeword size and number of codeword bytes allocated in step 3 with the codeword size obtained in step 4. Truncate the codeword size obtained in the encoded codeword at the allocated codeword size and discard the excess codewords.

[0033] Furthermore, step 6, which involves performing MQ encoding and outputting a compressed bitstream, specifically includes:

[0034] Step 6-1: During the MQ encoding process, initialize the interval length register, the lower bound register, the shift register, and the probability estimation index and high probability symbol for different context environments of each encoding channel.

[0035] Step 6-2: Input data stream. Determine whether the input data is a high-probability symbol based on the context information corresponding to each data in the bit stream. Perform corresponding probability encoding and output the encoded value in the register one by one to obtain the compressed bit stream. Clear the value in the register after the data stream ends.

[0036] Furthermore, if the symbol in step 6-2 is of high probability, then high probability encoding is performed. Specifically, the length of the high probability symbol interval is first assigned to register A. If A > 0x8000, the length of the high probability symbol interval is definitely greater than the length of the low probability interval, and no interval swapping is required. The value of the lower bound register is incremented by Qe, where Qe is the probability obtained from the table for the corresponding symbol. If the length of the high probability symbol interval is less than the length of the low probability symbol interval, the two intervals need to be swapped to ensure that the high probability symbol takes the larger interval. The value of register A is updated to Qe, the value of the lower bound register remains unchanged, and then the probability estimation index is updated, and the registers are renormalized.

[0037] If the symbol in step 6-2 is a low-probability symbol, then low-probability encoding is performed. Low-probability symbol encoding is similar to high-probability symbol encoding, but the difference is that after multiple low-probability symbol encodings, the high and low probability symbols may be swapped. It is necessary to check the current probability estimation table to determine whether a switch is needed.

[0038] Compared with the prior art, the significant advantages of this invention are:

[0039] (1) It can obtain code words in advance, reduce unnecessary encoding, and shorten encoding time.

[0040] (2) In bit-plane coding, the coding time is greatly shortened, while the context codewords are reduced, and the code stream transmission capacity is reduced.

[0041] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0042] Figure 1 This is a flowchart of the JPEG2000 algorithm based on channel truncation and bit plane scanning of the present invention.

[0043] Figure 2 This is a schematic diagram of wavelet transform.

[0044] Figure 3 This is a JPEG 200-bit planar encoded scan.

[0045] Figure 4 This is a one-time bit-plane encoding scan map.

[0046] Figure 5 for Figure 4 The problem diagram represented in the one-time bit-plane encoding.

[0047] Figure 6 This is a comparison chart of bit-plane coding performance.

[0048] Figure 7 The image is a comparison of the decoded images after 10x compression and truncation of the bit plane. Figure 7 (a) in the image is the original JPEG2000 decoded image. Figure 7 (b) in the figure is the image after JPEG2000 decoding according to the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0050] In one embodiment, combined Figure 1 This paper provides a JPEG2000 algorithm based on channel truncation and bit-plane scanning, the algorithm comprising the following steps:

[0051] Step 1: Input the image to be compressed;

[0052] Step 2: Perform wavelet transform on the input image and divide the resulting matrix into four sub-bands: LL, LH, HL, and HH.

[0053] Step 3: Assign codewords to each subband after wavelet transform;

[0054] Step 4: Perform bit-plane channel scanning encoding on the LL subband, and perform one-time bit-plane scanning encoding on the other subbands;

[0055] Step 5: Truncate the encoding result from Step 4;

[0056] Step 6: Perform MQ encoding and output the compressed bitstream.

[0057] Furthermore, in one embodiment, step 2 preferably employs a 5 / 3 wavelet transform, the forward transform formula of which is:

[0058]

[0059]

[0060] Its inverse transformation formula is:

[0061]

[0062]

[0063] In the formula, For the odd-numbered rows / columns of the input matrix, For the even-numbered rows / columns of the input matrix, For the next even row / column of the input matrix, For the odd-numbered rows / columns of the matrix obtained by wavelet transform, For an odd-numbered row / column of the resulting matrix, The even-numbered rows / columns of the resulting matrix;

[0064] The above formula is used to perform row and column transformations on the input image to complete the wavelet transform.

[0065] Furthermore, in one embodiment, the specific formula for codeword allocation in step 3 is as follows:

[0066]

[0067] in, It is the number of bytes allocated to the i-th subband. It is the target codeword count, which is the length that should be compressed and truncated. It is the number of encoded channels in the i-th sub-band. These are the wavelet subband weighting coefficients of the i-th subband.

[0068] in, =3*p-1, where p is the number of bit planes.

[0069] Furthermore, in one embodiment, step 4, which involves performing a one-time bit-plane scan encoding on the non-LL subband, specifically includes:

[0070] Step 4-1: Perform bit-plane scanning. In JPEG2000 bit-plane encoding, the scanning method involves scanning and judging in groups of four columns. Each judgment requires assigning a channel and different context codewords. The scanning method is as follows: Figure 3 As shown, a strip of four bits is used as the code block, and the code is scanned sequentially from left to right and from top to bottom until all coefficients in the code block are scanned.

[0071] Step 4-2: Perform one-time bit-plane encoding on each scanned column:

[0072] Step 4-2-1: Determine whether the coefficients of the current codeword sample to be encoded and its neighboring samples are all significant. If they are not significant, then perform run-length encoding, i.e., RLC encoding; otherwise, proceed to the next step.

[0073] Step 4-2-2: Determine whether the coefficient of the current codeword sample is significant. If it is significant, perform amplitude refinement encoding, i.e., MR encoding. Then determine whether the current column has been scanned. If the scan is complete, proceed to step 4-2-4; otherwise, repeat step 4-2 to continue encoding. If it is not significant, perform zero encoding, i.e., ZC encoding, and then proceed to the next step.

[0074] Step 4-2-3: Determine if the current codeword sample to be encoded is 0. If not, perform saliency encoding, i.e., SC encoding. Otherwise, determine if the current column has been scanned. If the scan has been completed, execute step 4-2-4. Otherwise, repeat step 4-2 to continue encoding.

[0075] Step 4-2-4: Determine whether the bit plane scanning is complete. If the scanning is complete, end the bit plane encoding; otherwise, start column scanning from step 4-1.

[0076] Here, the specific method for significance judgment in step 4-2-1 is as follows: if there is a bit '1' in the bit plane before the sample where the current encoded bit is located, then the significance of the sample where the encoded bit is located is '1', indicating significance; otherwise, it is '0', indicating insignificance.

[0077] Preferably, in step 4-2-1, the neighbor samples include the eight neighbor samples of the current codeword sample to be encoded.

[0078] Here, the encoding method is applicable to the LH, HL, and HH subbands. For the LL subband, which has rich information content, using one-time bit-plane encoding would actually increase the number of context codewords. The encoding flowchart is as follows. Figure 4 As shown. Figure 5 for Figure 4 The meaning of the letters represented in the text.

[0079] This part of the encoding essentially involves entropy encoding of various encoding elements, primarily using context-based binary arithmetic encoding. After wavelet transform and quantization, the slice component matrix becomes a sub-band matrix with integer coefficients. Each sub-band is further divided into rectangular code blocks of equal size, with the width and height of each block being integer powers of 2. Each code block can then be divided into bit planes, or bit layers, and encoding is performed plane by plane, starting from the most significant bit plane and proceeding down to the least significant bit plane.

[0080] This section encodes each bit plane using three channels and employs four encoding operations to encode the context information of the bit plane. Finally, it outputs an embedded bit stream, an array containing the encoded length information, and data containing the context information index. Specifically, for the highest non-all-zero bit plane, all bits belong to the clear channel. Bits in other bit planes can only belong to one of these channels. The channel decision is closely related to the saliency state of the sample. If a bit '1' exists in the bit plane preceding the sample containing the currently encoded bit, then the current saliency of the sample containing the encoded bit is '1'.

[0081] The three scanning processes are saliency propagation, amplitude refinement, and clearing. The combined encoding of all these processes constitutes the encoding after scanning the bit plane. Each process can be further subdivided into four encoding methods: zero encoding, sign encoding, amplitude refinement encoding, and run-length encoding. These four encoding methods yield the encoded bit stream (D) and context index (CX) information data of the bit plane.

[0082] In the original JPEG2000, bit-plane encoding was used to encode and allocate important information contained in the bit plane. Within the bit plane, the scanning method involved scanning and judging in groups of four columns, assigning each column to its corresponding channel and different context codewords each time. The scanning method was as follows: Figure 2 As shown, a strip of four bits is used as the code, and the code is scanned sequentially from left to right and from top to bottom until all coefficients in the code block are scanned. If a bit that does not belong to the current encoding channel is encountered during the scanning process, the encoding result is discarded. Bit-plane encoding consists of three channels and four encoding operations. The context and its corresponding codeword are obtained through the encoding operation of each channel. The scheme in this embodiment eliminates the need for scanning of different channels by performing a single bit-plane scan encoding. Each bit-plane is scanned only once, which reduces the encoding output and encoding time.

[0083] Furthermore, in one embodiment, step 5, based on the codeword allocation in step 3, truncates the encoding results of other subbands in step 4, specifically including:

[0084] Compare the codeword size and number of codeword bytes allocated in step 3 with the codeword size obtained in step 4. Truncate the codeword size obtained in the encoded codeword at the allocated codeword size and discard the excess codewords.

[0085] Here, the truncation is optimized in the original JPEG2000. Each bit-plane encoding channel is used as a truncation point, and the distortion at each truncation point is calculated. The truncation point is selected based on the compression ratio, truncating the bitstream while minimizing distortion and ensuring the length meets the compression ratio. Finally, the image is decoded. Calculating the truncation point after wavelet transform reduces the codewords in the bit-plane encoding before encoding and also effectively performs truncation.

[0086] In this invention, truncation based on the number of bit planes is combined with a one-time bit plane encoding method, which reduces the total encoding time and the length of the compressed bit stream, achieves fast encoding, and slightly improves compression performance.

[0087] Furthermore, in one embodiment, step 6, which involves performing MQ encoding and outputting a compressed bitstream, specifically includes:

[0088] Step 6-1: During the MQ encoding process, initialize the interval length register, the lower bound register, the shift register, and the probability estimation index and high probability symbol for different context environments of each encoding channel.

[0089] Step 6-2: Input data stream. Determine whether the input data is a high-probability symbol based on the context information corresponding to each data in the bit stream. Perform corresponding probability encoding and output the encoded value in the register one by one to obtain the compressed bit stream. Clear the value in the register after the data stream ends.

[0090] If the symbol is of high probability, high probability encoding is performed. Specifically, the length of the high probability symbol interval is first assigned to register A. If A > 0x8000, the length of the high probability symbol interval is definitely greater than the length of the low probability interval, and no interval swapping is required. The value of the lower bound register is incremented by Qe, where Qe is the probability obtained from the table for the corresponding symbol. If the length of the high probability symbol interval is less than the length of the low probability symbol interval, the two intervals need to be swapped to ensure that the high probability symbol takes the larger interval. The value of register A is updated to Qe, the value of the lower bound register remains unchanged, and then the probability estimation index is updated, and the registers are renormalized.

[0091] If it is a low-probability symbol, then low-probability encoding is performed. Low-probability symbol encoding is similar to high-probability symbol encoding, but the difference is that after multiple low-probability symbol encodings, the high and low probability symbols may be swapped. It is necessary to check the current probability estimation table to determine whether a switch is needed.

[0092] As a specific example, this embodiment further verifies and illustrates the invention. This embodiment compares the image compression performance of Lenna images, taking a 10x compression ratio of the codewords obtained after bit-plane encoding as an example.

[0093] Step 1, Calculate the cut-off length

[0094] The compressed bitstream length is calculated using the formula in equation (1), the total length of the input image is calculated, and then the codeword obtained by encoding is calculated to be one-tenth of the original length based on the compression ratio. The input is then used to obtain... =174490, and a three-level wavelet transform is used. The corresponding wavelet sub-band coefficients are LL3=0.67188, LH3=HL3=0.72992, HH3=0.79297, LH2=HL2=0.79611, HH2=0.92188, LH1=HL1=1.03833, HH1=1.43750. The number of channels of the input Lenna image is calculated according to the different sub-band bit planes. The number of channels of each sub-band is 3 times the number of bit planes and then 2 minus. The resulting number of channels for LL3, HH3, HL2, and HH2 is 22, the number of channels for LH3, HL3, HL1, and LH2 is 19, and the number of channels for LH1 and HH1 is 16. Finally, the codewords for each sub-band are obtained using equation (1): LL3=36035, LH3=16508, HL3=14214, HH3=7698, LH2=11701, HL2=12521, HH2=11377, LH1=7263, HL1=8265, HH1=4580. The LL3 sub-band code is encoded using the original algorithm.

[0095] Step 2: Perform wavelet transform

[0096] A 5 / 3 wavelet is used to perform a 3-level transform, and the image is processed into ten sub-bands as described above. Each sub-band is then individually bit-plane scan encoded.

[0097] Step 3: Perform a one-time bit-plane scan encoding.

[0098] First, it's important to note that because the LL subband contains a rich amount of information, using a single bit-plane scan for encoding would actually increase the codeword count. Therefore, this subband uses the original scanning method for bit-plane encoding, while the remaining subbands use a single scan to encode all bit planes. The scanning process for each column is as follows: Figure 2 As shown, the algorithm first determines whether there are any importance coefficients (i.e., whether there are any coefficients equal to 1) among the 3x3 neighboring coefficients of the given position. If not, it performs run-length encoding (RLC encoding), encoding the coefficients of a column using the corresponding context codewords. The LH1 subband is used as an example to demonstrate the difference in encoding speed and codeword count before and after encoding; the other subbands are similar. Figure 6 The encoding time of the algorithm refers to the encoding time in the bit plane. It can be seen that, under the condition that the code structure of the two is similar, the original algorithm takes 21.866s for bit plane encoding, while the improved algorithm takes 5.971s. This eliminates the need for multiple channel scans, and the improved codewords are reduced by about 11% compared to the original.

[0099] The subsequent steps are the same as the original algorithm, just normal encoding and decoding. The compression performance is not much different from the original algorithm, but the overall encoding speed is faster.

[0100] Step 4: Perform MQ encoding or other entropy encoding as in the original JPEG2000. If it needs to be encapsulated into a bitstream, encapsulate it according to the requirements of the original JPEG2000.

[0101] After decoding the original JPEG2000 compression algorithm and the new algorithm respectively, the recovered images are as follows: Figure 7 As shown, its performance is better than the original JPEG2000.

[0102] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention without departing from its spirit and scope should be included within the protection scope of the present invention.

Claims

1. A JPEG2000 method based on channel truncation and bit-plane scanning, characterized in that, The method includes the following steps: Step 1: Input the image to be compressed; Step 2: Perform wavelet transform on the input image and divide the resulting matrix into four sub-bands: LL, LH, HL, and HH. Step 3: Assign codewords to each subband after wavelet transform; Step 4: Perform bit-plane channel scanning encoding on the LL subband, and perform one-time bit-plane scanning encoding on the other subbands; Step 5: Based on the codeword allocation in Step 3, truncate the encoding results of other subbands in Step 4; Step 6: Perform MQ encoding and output the compressed bitstream; In step 2, the wavelet transform used is the 5 / 3 wavelet transform, and its forward transform formula is: Its inverse transformation formula is: In the formula, For the odd-numbered rows / columns of the input matrix, For the even-numbered rows / columns of the input matrix, For the next even row / column of the input matrix, For the odd-numbered rows / columns of the matrix obtained by wavelet transform, For an odd-numbered row / column of the resulting matrix, The even-numbered rows / columns of the resulting matrix; The above formula is used to perform row and column transformations on the input image to complete the wavelet transform. Step 4, which involves performing a one-time bit-plane scan encoding on other sub-bands, specifically includes: Step 4-1: Perform bit plane scanning until all coefficients in the code block have been scanned; Step 4-2: Perform one-time bit-plane encoding on each scanned column: Step 4-2-1: Determine whether the coefficients of the current codeword sample to be encoded and its neighboring samples are all significant. If they are not significant, then perform run-length encoding, i.e., RLC encoding; otherwise, proceed to the next step. Step 4-2-2: Determine whether the coefficient of the current codeword sample is significant. If it is significant, perform amplitude refinement encoding, i.e., MR encoding. Then determine whether the current column has been scanned. If the scan is complete, proceed to step 4-2-4; otherwise, repeat step 4-2 to continue encoding. If it is not significant, perform zero encoding, i.e., ZC encoding, and then proceed to the next step. Step 4-2-3: Determine if the current codeword sample to be encoded is 0. If not, perform saliency encoding, i.e., SC encoding. Otherwise, determine if the current column has been scanned. If the scan has been completed, execute step 4-2-4. Otherwise, repeat step 4-2 to continue encoding. Step 4-2-4: Determine whether the bit plane scanning is complete. If the scanning is complete, end the bit plane encoding; otherwise, start column scanning from step 4-1.

2. The JPEG2000 method based on channel truncation and bit-plane scanning according to claim 1, characterized in that, The specific formula for codeword allocation in step 3 is as follows: in, It is the number of bytes allocated to the i-th subband. It is the target codeword count, which is the length that should be compressed and truncated. It is the number of encoded channels in the i-th sub-band. These are the wavelet subband weighting coefficients of the i-th subband.

3. The JPEG2000 method based on channel truncation and bit-plane scanning according to claim 2, characterized in that, The =3*p-1, where p is the number of bit planes.

4. The JPEG2000 method based on channel truncation and bit-plane scanning according to claim 1, characterized in that, The specific method for significance determination in step 4-2-1 is as follows: If there is a bit '1' in the bit plane before the sample where the current encoded bit is located, then the significance of the sample where the encoded bit is located is '1', indicating significance; otherwise, it is '0', indicating insignificance.

5. The JPEG2000 method based on channel truncation and bit-plane scanning according to claim 1, characterized in that, In step 4-2-1, the neighbor samples include the eight neighbor samples of the current codeword sample to be encoded.

6. The JPEG2000 method based on channel truncation and bit-plane scanning according to claim 4, characterized in that, Step 5, based on the codeword allocation in step 3, involves truncating the encoding results of other subbands in step 4. Specifically, this includes: Compare the codeword size and number of codeword bytes allocated in step 3 with the codeword size obtained in step 4. Truncate the codeword size obtained in the encoded codeword at the allocated codeword size and discard the excess codewords.

7. The JPEG2000 method based on channel truncation and bit-plane scanning according to claim 6, characterized in that, Step 6, which involves performing MQ encoding and outputting a compressed bitstream, specifically includes: Step 6-1: During the MQ encoding process, initialize the interval length register, the lower bound register, the shift register, and the probability estimation index and high probability symbol for different context environments of each encoding channel. Step 6-2: Input data stream. Determine whether the input data is a high-probability symbol based on the context information corresponding to each data in the bit stream. Perform corresponding probability encoding and output the encoded value in the register one by one to obtain the compressed bit stream. Clear the value in the register after the data stream ends.

8. The JPEG2000 method based on channel truncation and bit-plane scanning according to claim 7, characterized in that, If the symbol in step 6-2 is of high probability, then high probability encoding is performed. Specifically, the length of the high probability symbol interval is first assigned to register A. If A > 0x8000, the length of the high probability symbol interval is definitely greater than the length of the low probability interval, and no interval swapping is needed. The value of the lower bound register is incremented by Qe, where Qe is the probability obtained from the table lookup for the corresponding symbol. If the length of the high probability symbol interval is less than the length of the low probability symbol interval, the two intervals need to be swapped to ensure that the high probability symbol takes the larger interval. The value of register A is updated to Qe, the value of the lower bound register remains unchanged, and then the probability estimation index is updated, and the registers are renormalized. If the symbol in step 6-2 is a low-probability symbol, then low-probability encoding is performed. Low-probability symbol encoding is similar to high-probability symbol encoding, but the difference is that after multiple low-probability symbol encodings, the high and low probability symbols may be swapped. It is necessary to check the current probability estimation table to determine whether a switch is needed.

Images