Image Processing Feature Extraction for Mixed-Class Region Distribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation methods struggle with accurately classifying regions where multiple classes are mixed, such as sky visible through tree branches, due to similar textures, leading to incorrect classification as either foreground or sky.
Innovation Solution
An image processing apparatus and method that estimates the distribution of regions with different attributes by extracting feature amounts and using supervisory information to train an estimator, considering the distribution-related information of classes within a region.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If region segmentation methods are used to classify regions into classes, then classification capability is provided, but accuracy deteriorates when multiple classes are mixed with similar textures
Solution Approach 1:
The patent introduces a new dimension of analysis by estimating the distribution of classes within each region rather than simply classifying regions into discrete categories. This distribution-based approach captures the compositional information of mixed classes, allowing the system to distinguish between regions with similar textures but different class compositions, thereby improving classification accuracy while maintaining versatility.
Solution Approach 2:
The patent changes the parameter representation from discrete class labels to continuous distribution parameters. By representing class composition as distribution parameters (e.g., proportions of different classes within a region), the system can capture subtle variations in texture and composition that discrete classification cannot detect, leading to improved accuracy in mixed-class scenarios.
2Productivity
If deep learning methods are used for semantic segmentation, then processing capability is enhanced, but computational complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the image into multiple regions and analyzing the class distribution within each region independently. This regional segmentation approach allows the system to leverage deep learning capabilities for feature extraction while maintaining manageable computational complexity through localized processing. The distribution estimation is performed per region rather than requiring complex global optimization.
Data Source
AI summary
There is provided with an image processing apparatus, for example, for image recognition. An extraction unit extracts a feature amount from a target image. An estimation unit estimates distribution of regions having attributes different from each other in the target image based on the feature amount.