TWI396448B - Entropy decoding device - Google Patents

Description

Entropy decoding device

The present invention relates to decoding, and more particularly to an entropy decoding apparatus.

Context-based adaptive binary arithmetic code (CABAC) is one of the tools for H.264/AVC compressed video data bitstream. The entropy decoding process of CABAC is the traversing of the binary tree formed by the stream of video data bits. However, the throughput of entropy decoding is often limited by complex algorithm calculations and dependencies between binary tree nodes.

When H.264/AVC is used for data processing, the entropy decoding part accounts for a large part of the calculation amount. If the speed of entropy decoding can be improved, the speed of H.264/AVC video decoding will be directly improved.

In view of the above, it is desirable to provide an entropy decoding apparatus for improving entropy decoding speed.

The entropy decoding device provided in the embodiment of the present invention includes a data storage module, a lookup table module, a prediction module, an algorithm decoding module, an update module, an output module, and a state register. The data storage module is configured to store a stream of video data bits to be decoded. The query table module is used to select a context probability model, which includes all nodes of the decoding tree corresponding to the video data bit stream and a most probable symbol (MPS) of each node. The prediction module is used to predict the next node information based on the current node and the MPS. The algorithm decoding module is configured to decode the video data bit stream according to the current node, and the decoded content includes the next node information. . The update module is configured to determine whether the next node information in the decoded content matches the predicted next node information, and when not matched, generates a push instruction to the lookup table module, and updates the MPS in the context probability model according to the push instruction. The output module is used to output decoded content. The state register is used to read and temporarily store the current node and the most likely symbol in the query table module, and output to the prediction module and the algorithm decoding module.

The above and the technical effects of the invention can be easily understood from the following detailed description of the embodiments and the accompanying drawings.

100‧‧‧Entropy decoding device

10‧‧‧Reader

11‧‧‧Data Storage Module

12‧‧‧Enquiry Form Module

120‧‧‧Context table

13‧‧‧Status register

20‧‧‧Executive

21‧‧‧ algorithm decoding module

22‧‧‧Analytical Module

23‧‧‧ Forecasting Module

24‧‧‧Update Module

25‧‧‧Timer

30‧‧‧ Output

31‧‧‧Output module

1 is a block diagram of an entropy decoding apparatus in an embodiment of the present invention.

2 is a flow chart of a method of entropy decoding in an embodiment of the present invention.

3 is a graph of prediction and decoding tree traversal in an embodiment of the present invention.

1 is a block diagram of an entropy decoding apparatus 100 according to an embodiment of the present invention.

In the present embodiment, the entropy decoding apparatus 100 is configured to decode context-based adaptive binary arithmetic code (CABAC) video data in accordance with the H.264/AVC standard. The H.264/AVC standard specifies a detailed decoding table and decoding process. Each decoding table corresponds to a decoding tree of a binary tree structure, and each decoding tree includes a plurality of branch nodes and a plurality of leaf nodes. Entropy decoding is to determine the path to access the decoding tree based on the video data, and read the value stored in the leaf node according to the path. In other embodiments, the entropy decoding apparatus 100 can also be used to process other video coding and decoding standards that are compatible with the H.264/AVC standard, for example, context-based adaptive variable length coding (CAVLC). )Wait.

In the present embodiment, the entropy decoding apparatus 100 includes a read end 10, an execution end 20, and an output end 30. The read terminal 10 includes a data storage module 11, a table look-up module 12, and a state buffer 13.

The data storage module 11 is configured to store an encoded bit stream to be decoded.

The query table module 12 is configured to select a context probability model, and provide a dependency relationship between nodes in the decoding tree corresponding to the video data bit stream to be decoded.

In the present embodiment, the lookup table module 12 includes a context table 120. Context table 120 is used to store a variety of context probability models as specified in H.264/AVC.

The context probability model includes a most probable symbol (MPS), a least probable symbol (LPS), and a state value for each branch node in the decoding tree of the decoding tree. In the present embodiment, the value of the MPS is derived from the context probability model specification of the H.264/AVC standard, and is used to indicate whether the next node is more likely to be left or right, and is generally represented by 0 and 1. Among them, MPS=0 is used to indicate that the probability to the left is large, and MPS=1 is used to indicate that the probability to the right is large. In the present embodiment, the context probability model is position dependent, that is, there are different MPS values at different locations.

The status register 13 is used to read and temporarily store the current node and the MPS in the lookup table module 12. In the present embodiment, the state register 13 performs a first-in first-out (FIFO) read and discard of the current node and the MPS value by means of a pipeline. In the present embodiment, the state register 13 adopts a pipeline method to enhance flexibility in reading and discarding nodes to improve the processing speed of the entropy decoding apparatus 100.

The execution terminal 20 includes an algorithm decoding module 21, an analysis module 22, a prediction module 23, an update module 24, and a timer 25. Timer 25 is used to provide a clock cycle.

The algorithm decoding module 21 is connected to the state register 13 and the data storage module 11 for reading the state. The current node in the register 13 decodes the video data bit stream in the data storage module 11, for example, a binary decode symbol (BDS), and outputs the decoded content. In the present embodiment, the decoded content includes information of the next node.

The parsing module 22 is configured to analyze whether the current node decoded by the algorithm decoding module 21 is a leaf node or a branch node. In the present embodiment, when the current node is a leaf node, the parsing module 22 outputs the decoded content of the leaf node to the output terminal 30. When the current node is a branch node, the parsing module 22 outputs the decoded content of the branch node to the update module 24.

The prediction module 23 is used to predict the next node information. In the present embodiment, the prediction module 23 predicts the next node as the left or right child of the current node according to the current node in the state register 13 and the MPS.

In the present embodiment, the algorithm decoding module 21 performs the prediction of the current node and the prediction of the next node by the prediction module 23 in the same clock cycle of the timer 25.

The update module 24 is connected to the analysis module 22 and the prediction module 23, and is configured to determine whether the information of the next node in the decoded content of the branch node matches the predicted next node information, and when the prediction does not match (prediction miss When a flush _[a1] instruction is generated and output to the lookup table module 12, and the MPS in the context table 120 is updated according to the push instruction.

In the present embodiment, the state register 13 is further configured to discard the predicted next node information after the lookup table module 12 receives the push instruction, and re-read the next node information included in the decoded content.

In the present embodiment, the algorithm decoding module 21 is further configured to read the next node information in the state register 13 for decoding when a prediction hit is predicted.

In the present embodiment, the lookup table module 12 updates the MPS after receiving the push command. In the present embodiment, the update MPS is based on the probability state transition mode specified by H.264/AVC. The probabilistic model is maintained, including maintaining the offset of the current probability interval (offset) and the current interval width (range).

The entropy decoding device 10 can change the direction of subsequent prediction by updating the MPS. For example, from MPS=0, it is converted to MPS=1.

The output terminal 30 includes an output module 31 for outputting decoded content of the leaf node.

Please refer to FIG. 2, which is a flowchart of a decoding method in an embodiment of the present invention.

In step S201, the data storage module 11 stores the video material bit stream to be decoded.

In step S202, the lookup table module 12 selects a context probability model corresponding to the video material bit stream to be decoded in the context table 120, including decoding all nodes in the tree and the MPS of each node. In the present embodiment, the MPS of the leaf node is null.

In step S203, the status register 13 reads the current node and the MPS from the lookup table module 12. In the present embodiment, the state register 13 performs a first-in first-out (FIFO) read and discard of the current node and the MPS by means of a pipeline.

In step S204, the prediction module 23 predicts the next node information, and the algorithm decoding module 21 receives the current node, decodes the video data bit stream in the data storage module 11, and outputs the decoded content.

In the present embodiment, the prediction module 23 and the algorithm decoding module 21 perform prediction and decoding, respectively, in the same clock cycle. In the present embodiment, the prediction module 23 predicts the next node as the left or right child of the current node according to the current node in the state register 13 and the MPS.

In step S205, the parsing module 22 is configured to analyze whether the current node decoded by the algorithm decoding module 21 is a leaf node or a branch node.

If the current node is a leaf node, step S206 is performed, and the parsing module 22 decodes the leaf node. The content is output to the output module 31, and the output content of the leaf node is output by the output module 31.

If the current node is a branch node, step S207 is performed, and the update module 24 determines whether the information of the next node in the decoded content of the branch node matches the predicted next node.

If they match, the process returns to step S203. In the present embodiment, the current node is the node that has been updated according to the prediction.

If not, step S208 is executed, and the update module 24 generates a push command output to the lookup table module 12, discards the predicted next node information from the state register 13, and returns to step S202. In the present embodiment, after receiving the push-out instruction, the look-up table module 12 updates the MPS according to the branch decoding content. In the present embodiment, the state register 13 performs the reading and discarding of steps S203 and S208 in a pipeline manner. In the present embodiment, the MPS is updated, and the context probability model is maintained in accordance with the probability state transition method defined by H.264/AVC.

The prediction and traversal of the decoding tree are shown in Figures 3(1) and (2). In FIG. 3(2), the abscissa indicates the clock period, the ordinate indicates the node, F indicates the read end 10 of the entropy decoding apparatus 100, E indicates the execution end 20 of the entropy decoding apparatus 100, and O indicates the output of the entropy decoding apparatus 100. End 30.

The lookup table module 12 selects a decoding tree, as shown in FIG. 3(1), the current node is 0, and the MPS is 0.

As shown in Fig. 3(2), when the decoding coincides with the prediction, if the current node is 0, the next node 2 is continuously read, and decoding and prediction are continued.

If they do not match, if the current node is 2, the predicted next node 3 is discarded, the next node 4 is read, and the MPS is updated. According to the MPS prediction of the next node, the processing speed of the node can be improved. When the MPS is continuously updated according to the decoded content, the accuracy of the prediction can be improved, and the speed of the entropy decoding apparatus 100 in the invention can be improved.

In summary, the present invention complies with the requirements of the invention patent and submits a patent application according to law. However, the above The present invention is only a preferred embodiment of the present invention, and equivalent modifications or variations made by those skilled in the art will be included in the following claims.

100‧‧‧Entropy decoding device

10‧‧‧Reader

11‧‧‧Data Storage Module

12‧‧‧Enquiry Form Module

120‧‧‧Context table

13‧‧‧Status register

20‧‧‧Executive

21‧‧‧ algorithm decoding module

22‧‧‧Analytical Module

23‧‧‧ Forecasting Module

24‧‧‧Update Module

25‧‧‧Timer

30‧‧‧ Output

31‧‧‧Output module

Claims (8)

An entropy decoding device, comprising: a data storage module, configured to store a video data bit stream to be decoded; and a query table module, configured to select a context probability model, where the context probability model includes the video data bit stream corresponding to All nodes of the decoding tree and the most likely symbols of each node; a state register for reading and temporarily storing the current node in the query table module and the most likely symbol; prediction mode a group, configured to predict a next node information according to the current node outputted by the state register and the most likely symbol; an algorithm decoding module, configured to output the current node pair according to the state register The video data bit stream is decoded, and the decoded content includes the next node information; and the update module is configured to determine whether the next node information in the decoded content matches the predicted next node information, and When not matching, a push instruction is generated to the lookup table module, and the most likely symbol in the context probability model is updated according to the push instruction; and an output module is configured to output the decoded content. The entropy decoding device of claim 1, wherein the status register is further configured to discard the current node after receiving the push instruction by the lookup table module, and according to the next in the decoded content. The node information is re-read. The entropy decoding apparatus according to claim 1, further comprising a timer for providing a clock cycle, wherein the decoding operation of the algorithm decoding module and the prediction operation of the prediction module are performed in the same clock cycle. The entropy decoding device of claim 1, further comprising an analysis module for parsing the The current node is a leaf node or a branch node. The entropy decoding device of claim 4, wherein the parsing module is further configured to output the decoded content of the leaf node to the output module when the current node is the leaf node. The entropy decoding device of claim 4, wherein the parsing module is further configured to output the decoded content of the branch node to the update module when the current node is the branch node. The entropy decoding device of claim 1, wherein the query table module further comprises a context table for storing the context probability model. The entropy decoding apparatus according to claim 1, wherein the update is most likely to occur, and the context probability model is maintained according to a probability state transition manner specified by H.264/AVC.