Entropic Transcoding Subword Segmentation for Noise-Resilient Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing entropic codes struggle to efficiently compress binary data from real-world signals like images and videos due to noise in low-order bits, which disrupts the statistical probability decay of coefficients, reducing compression potential.
Innovation Solution
A method for entropic transcoding that subdivides each word into low-order and high-order subwords, where the entropic code is applied only to the high-order subwords, with the low-order bits considered as noise, optimizing the encoding process by excluding them from the encoding process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If entropic code is applied to all bits of binary data, then compression efficiency is maximized for ideal statistical models, but compression performance deteriorates when noise is present in low-order bits
Solution Approach 1:
The binary data is segmented into two parts: significant bits (high-order) and noise bits (low-order). The entropic code is applied only to the significant bits, while noise bits are handled separately or discarded. This segmentation resolves the contradiction by applying compression only where statistical regularities exist, avoiding the negative impact of noise on compression efficiency.
2Quantity of substance
If entropic code uses variable-length encoding for all coefficients, then compression ratio is improved, but encoding complexity increases
Solution Approach 1:
The encoding process is segmented into two stages: first, determine the position of the most significant bit; second, apply variable-length encoding only to the significant portion. This reduces the overall encoding complexity compared to applying variable-length encoding to all bits, while still achieving good compression ratios.
3Ease of manufacture
If fixed-size binary words are used for all coefficients, then encoding simplicity is maintained, but compression potential is lost
Solution Approach 1:
Different encoding strategies are applied to different parts of the data: fixed-size encoding for noise bits (simple) and variable-length encoding for significant bits (compressive). This local differentiation resolves the contradiction by maintaining simplicity where needed while achieving compression where possible.
Data Source
AI summary
A method for entropic transcoding of a first sequence of binary data of words to transcode into a second compressed sequence of binary data uses a predetermined entropic code involving a variable-length of the encoded words to transcode each word of the first sequence of binary data into a transcoded word. Based on a predetermined number B of low-order bits, the method includes application to each word of the first sequence of binary data: subdivision of the word into first and second subwords, wherein the first subword includes the B low-order bits of the word, and the second subword includes the other high-order bits of the word, application of the predetermined entropic code to the second subword to obtain a second transcoded subword, and obtaining the transcoded word by concatenation of the first subword and of the second transcoded subword.