Encoding method, decoding method and application based on huffman coding

Abstract

The invention discloses an encoding method based on Huffman encoding. The encoding method comprises the following steps: Huffman encoding the input sequence with a first designated length as an encoding unit; and a lengthening step, for obtaining a Huffman encoding sequence composed of encoded values from a Huffman encoding module, as the boundary is a second designate length, if the last encodedvalue input cannot be placed within a W-length independent sequence, the encoded value is put into the next independent sequence of the second specified length, and the remaining length in the currentindependent sequence is filled in with a bit-occupying code that does not affect the decoding result, and finally the encoded sequence composed of a plurality of independent sequences of the second specified length is sequentially connected is outputted. The invention also discloses a corresponding coding device based on Huffman coding, and a compression decompression method and a requirement ina convolutional neural network. The invention has the effect that boundary searching is not necessary, and the decoding efficiency is greatly improved.

Description

Background technique [0001] Huffman coding is a lossless entropy coding, which is widely used in information compression such as digital audio and video. [0002] Huffman coding is to classify the input sequence according to the probability. The higher probability is encoded with fewer bits, and the lower probability is encoded with more bits. When decoding, a coding mapping table is needed, and each coding value is searched in the table to restore the original data sequence. Such an encoding method can obtain a better compression ratio when there is more redundancy in the data sequence. [0003] Taking binary Huffman coding as an example, assuming that there are many identical values ​​in the storage sequence, we can use the Huffman coding principle to divide the data into two categories: one is the constant value MFV (most frequently value), generally The storage matrix is ​​searched to find the most frequently occurring value. This type of value will be mapped to the code...

AI Technical Summary

Problems solved by technology

[0014] However, high-speed transmission and processing systems are implemented in pipelines, and no matter how fast the Huffman boundary code is, it is difficult to keep up with the speed of the processor or master.

In addition, when the decoding layer timing of a Huffman decoder does not meet the system bandwidth requirements, a pipeline design is usually introduced, and the delay introduced thereby will cause address jumps When the pipeline is cleared and the address is "rolled back", if there is no predictable boundary, it is almost impossible to roll back or recalculate the address. Even if the codec is used in the prior art, because the Huffman code is of variable length , so the length of each independent data sequence is also uncertain. Even if the boundary code is used, the boundary is determined by searching, and the boundary is still unpredictable

Images