Image Coding Sign Prediction via Linear Pattern Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image compression methods require complex reconstruction processes for predicting sign coefficients, leading to high computational complexity due to the need for multiple inverse transformations and coefficient recalculations.
Innovation Solution
The method simplifies the prediction process by reusing results from previous assumptions, constructing pixel vectors through linear combinations of previously processed pixel vectors and pattern vectors, reducing the number of transformations needed, and applying a cost function calculation only to necessary pixels, thereby minimizing computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional reconstruction processes are used for predicting sign coefficients, then prediction accuracy is maintained, but computational complexity increases exponentially due to multiple inverse transformations and coefficient recalculations
Solution Approach 1:
The patent pre-calculates and stores pattern vectors for different sign combinations before actual decoding. These pattern vectors represent the contribution of each coefficient to the pixel values, allowing the decoder to quickly assemble pixel vectors by linear combination without performing full inverse transformations during decoding, thus reducing computational complexity while maintaining prediction accuracy
Solution Approach 2:
The patent changes the representation form from storing complete coefficient sets to storing pattern vectors that capture the essential contribution patterns. By representing coefficient contributions as pre-computed pattern vectors and combining them linearly, the system transforms the computational problem from exponential complexity (evaluating all 2^M sign combinations with full reconstruction) to linear complexity (combining M pattern vectors)
2Measurement precision
If all 2^M combinations of sign values are evaluated for prediction, then prediction accuracy is improved, but the number of transformations and calculations increases exponentially
Solution Approach 1:
The patent pre-computes pattern vectors that represent the pixel contributions of individual coefficients for different sign values. These pattern vectors are stored and reused during decoding, eliminating the need to perform full inverse transformations for each of the 2^M sign combinations, thus dramatically improving processing speed while still enabling accurate prediction through linear combination of relevant patterns
Solution Approach 2:
The patent transforms the problem from evaluating complete coefficient combinations to selecting and combining individual pattern vectors based on predicted signs. This parameter change from full reconstruction to selective linear combination reduces the computational burden from exponential to linear in the number of coefficients M, significantly improving processing speed
3Device complexity
If pattern vectors are pre-calculated and stored, then computational complexity during decoding is reduced, but memory requirements increase
Solution Approach 1:
The patent changes the stored representation from complete coefficient data to compact pattern vectors that capture only the essential contribution patterns. Each pattern vector represents how a single coefficient affects pixel values, allowing efficient linear combination during decoding. This parameter change significantly reduces the amount of data that needs to be stored and processed compared to storing complete coefficient sets for all possible sign combinations
Data Source
AI summary
A method for coding a digital image divided into blocks. The method includes, for a current block: processing the current block to provide a vector of transformed coefficients; selecting a sequence of M signs of coefficients to be predicted in the coefficient vector, M being an integer greater than or equal to 2; predicting values of the signs from a plurality of assumptions of combinations and a cost function; and for a sign of the sequence, coding an indicator representing a difference between its actual value and its predicted value. The predicting includes, for at least one second, current assumption: obtaining a pixel vector associated with the current assumption, by linear combination of at least one pixel vector associated with a first assumption and at least one pattern vector obtained from the vector of transformed coefficients; an calculating the cost function for the current assumption from the pixel vector obtained.