Matching Pursuit Image Compression for Real-Time Low Bit Rates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The practical application of the matching pursuits algorithm in video and image compression is limited due to its computationally intensive nature, making it unsuitable for real-time video coding systems without significant optimization or hardware speed increases.
Innovation Solution
A method of data compression that applies a transform to multi-dimensional data, using a plurality of one-dimensional matching pursuits algorithms in different scanning directions to generate and code a multi-dimensional transform data set, convolving the data with basis functions to identify and subtract atoms, and iteratively updating the data set to produce quantized transform data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If the matching pursuits algorithm is applied to multi-dimensional video data using conventional methods, then compression effectiveness is improved, but computational complexity increases making real-time processing impossible
Solution Approach 1:
The patent segments the multi-dimensional matching pursuits problem into multiple one-dimensional matching pursuits problems by applying transforms along different dimensions separately. Instead of performing 2D or 3D matching pursuits directly on video data, the invention applies 1D matching pursuits along horizontal and vertical dimensions independently after appropriate transforms, dramatically reducing computational complexity while maintaining compression effectiveness.
2Productivity
If the matching pursuits algorithm is applied to video data in real-time, then processing speed is improved, but compression quality deteriorates due to insufficient computation time
Solution Approach 1:
By segmenting the computation into separate 1D passes along different dimensions, the invention enables real-time processing while maintaining compression quality. Each 1D pass can be completed quickly, and the cumulative effect of multiple 1D passes approximates the quality of full multi-dimensional matching pursuits.
Solution Approach 2:
The patent transforms the problem from multi-dimensional matching pursuits to a sequence of one-dimensional operations by introducing dimensional separation through transforms. This dimensionality reduction allows real-time processing while preserving the essential compression benefits through iterative application along different dimensional axes.
3Loss of information
If conventional matching pursuits is used for video coding, then compression performance is improved, but the requirement for motion vectors and complex hardware increases
Solution Approach 1:
The invention eliminates the need for complex motion vector hardware by segmenting the matching pursuits operation into simple 1D passes. This dimensional separation removes the requirement for sophisticated motion estimation hardware, making the system more suitable for real-time implementation with simpler hardware architectures.
Data Source
AI summary
A method and apparatus for data compression comprises applying a decorrelating transform to multi-dimensional data to be compressed, then using a sequence of one or more one-dimensional matching pursuits algorithms to code the output of the transform. The invention finds particular application in video and still image coders, particularly real-time coders, those intended for use at very low bit rates, and those for which scalable bit rates are of importance.