Pixel Scatterness Evaluation for Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for image categorization based on color distributions are inadequate as they fail to distinguish between different shapes and variations, and cannot evaluate pixel distribution at various resolution scales, especially when pixels do not form connected groups.
Innovation Solution
A method that quantifies pixel characteristics by measuring scatterness, allowing for the classification of images based on pixel dispersion even if pixels do not form connected regions, and provides a scatterness sequence that reflects pixel distribution at different resolution scales through merging pixel blocks into single pixels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If color histogram-based categorization is used, then image classification can be performed, but images with similar color distributions but different spatial arrangements cannot be distinguished
Solution Approach 1:
The patent segments the image into multiple blocks and analyzes pixel distribution within each block separately. By dividing the image into regions and computing scatterness for each block, the method captures spatial arrangement information that global color histograms miss, enabling distinction between images with similar color distributions but different spatial patterns
Solution Approach 2:
The patent adds a spatial dimension to traditional color-based analysis by computing scatterness values that reflect both color and positional information. This transforms the classification approach from purely color-space analysis to a combined color-spatial analysis, enabling discrimination based on spatial coherence
2Measurement precision
If connected components analysis is used to evaluate pixel distribution, then spatial coherence can be assessed, but the method fails when pixels do not form connected groups
Solution Approach 1:
The patent changes the evaluation parameter from connectivity-based metrics to scatterness-based metrics that do not require connected components. By using scatterness which measures dispersion around a reference point, the method remains effective whether pixels are connected or disconnected, greatly increasing adaptability while maintaining spatial coherence evaluation capability
3Device complexity
If fixed-resolution analysis is used, then processing is simplified, but pixel distribution cannot be evaluated at different resolution scales
Solution Approach 1:
The patent implements a dynamic analysis approach where the image is processed at multiple resolution scales by varying the block size. The scatterness computation adapts to different scales by adjusting the block dimensions, allowing the same algorithm to evaluate pixel distribution at fine, medium, and coarse levels without requiring separate fixed-resolution analyses
Data Source
AI summary
A method of evaluating element distribution within a data set, such as pixel characteristics whose scatterness is quantified. Scatterness is evaluated for individual scale levels as the image is merged through smaller scale levels. At each level, every pixel block (i.e., 2×2) in the region of interest merges to one pixel to form the next level. Select characteristics of the pixels are assessed in the merge, and the image may be simplified to fewer pixel states so the merges require less processor overhead. In one merge mode if any pixel in the block has the desired characteristic, then the resulting merged pixel is set with that characteristic. The merging rate of the pixels is used to determine the scatterness for the pixel distribution. A scatterness sequence can be used separately or in combination with known techniques for comparing images in applications such as search and classification.