Image Block Clustering for Texture Homogeneity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image compression techniques, such as H.264/MPEG-4 AVC, are not optimal for compressing images with regions of homogeneous texture, as they fail to exploit texture similarity between disjointed areas and require transmission of a segmentation map that is costly in terms of memory space and non-deterministic.
Innovation Solution
A method that groups blocks with similar texture or color into clusters, allowing for implicit or refinement-based encoding of block parameters, eliminating the need for a segmentation map and enabling efficient pooling of data within these clusters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If region-based coding techniques are used to segment images into areas of homogeneous texture, then compression efficiency is improved, but memory space requirements increase due to transmission of segmentation map
Solution Approach 1:
The invention extracts and transmits only the essential cluster identification data and characteristic values needed for decoding, rather than transmitting complete segmentation maps with boundary information. This selective extraction reduces the quantity of transmitted data while maintaining compression efficiency benefits.
Solution Approach 2:
The invention applies different coding strategies to different regions by identifying clusters of blocks with similar characteristics, allowing localized optimization without requiring global segmentation map transmission. Each cluster is treated as a local unit with its own characteristic values.
2Productivity
If segmentation into regions of arbitrary shapes is performed, then texture similarity exploitation is improved, but determinism and reliability decrease
Solution Approach 1:
The invention segments the image into clusters of blocks based on characteristic value similarity, creating regions of homogeneous texture without requiring arbitrary shape boundaries. This segmentation approach maintains determinism by using objective similarity criteria rather than subjective region definitions.
Solution Approach 2:
The invention changes the parameter used for region definition from arbitrary geometric boundaries to objective characteristic value similarity (such as texture or color parameters). This parameter change enables deterministic clustering that reliably exploits texture similarity while avoiding the non-determinism of arbitrary segmentation.
3Device complexity
If causal spatial prediction is used between neighboring macroblocks, then decoding complexity is reduced, but compression efficiency deteriorates for disjointed areas of same texture
Solution Approach 1:
The invention introduces cluster identification data as an intermediary that enables distant blocks with similar characteristics to be grouped together. This intermediary allows the decoder to efficiently identify and apply characteristic values from reference clusters without requiring complex searches through all previously decoded blocks.
Solution Approach 2:
The invention creates a universal mechanism for predicting block characteristics that works both for causally connected neighboring blocks and for disjointed blocks with similar texture. The cluster-based approach unifies these two cases under a single prediction framework, improving compression efficiency without significantly increasing decoding complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for encoding an image or image sequence generating a data flow including data representative of pixel groups, or blocks, in one of said images (IE). The method comprises the steps of: grouping (C1) the blocks in a block cluster based on the proximity of their respective values corresponding to at least one parameter of a block to be encoded; determining (C2) a value of said parameter, said value being characteristic of said block cluster; encoding (C3) blocks of said cluster, wherein the values of said blocks for said parameter are implicitly encoded by inheritance of said characteristic value, or are encoded as refinements relative to said characteristic value, and encoding a data structure associated with said block cluster, said data structure including a piece of data associated with said characteristic value.