Image Region Extraction Using Spatial Change Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image region extraction techniques face challenges in accurately extracting target regions with large pixel value changes inside and small changes at boundaries, or where pixel values are similar to other parts and distance changes are significant, due to reliance on pixel value and distance differences.
Innovation Solution
An image region extracting apparatus and method that acquire pixel value and distance change patterns to extract target regions based on similarities between the designation region and image parts, using spatial change characteristics of pixel values and distances.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If target region extraction is based on difference between pixel values of adjacent pixels, then extraction can be performed automatically, but extraction accuracy deteriorates for regions with large pixel value changes inside or small changes at boundaries
Solution Approach 1:
The patent introduces a new dimension of analysis by examining spatial change characteristics (patterns of change) rather than only absolute pixel value differences. This transforms the extraction criterion from a one-dimensional pixel value comparison to a two-dimensional spatial pattern analysis, enabling accurate extraction of regions with complex pixel value distributions.
Solution Approach 2:
The patent changes the extraction parameter from absolute pixel value difference to spatial change characteristic similarity. By analyzing how pixel values change across space (spatial change characteristics) rather than their absolute differences, the system can accurately identify target regions regardless of their specific pixel value ranges or boundary characteristics.
2Extent of automation
If target region extraction is based on difference in distance from reference position, then extraction can be performed automatically, but extraction accuracy deteriorates for regions with large distance changes inside or small changes at boundaries
Solution Approach 1:
The patent extends the analysis from one-dimensional distance comparison to two-dimensional spatial pattern analysis by examining how distance changes across the image. This spatial change characteristic analysis enables accurate extraction of regions with complex distance distributions, such as objects at varying depths or regions with non-uniform depth gradients.
Solution Approach 2:
The patent transforms the extraction criterion from absolute distance difference to spatial change characteristic similarity. By analyzing the pattern of distance changes rather than absolute distance values, the system can accurately identify target regions with complex depth characteristics while maintaining automatic operation.
3Extent of automation
If target region extraction is based on similarity of pixel value component and distance component, then extraction can be performed automatically, but extraction accuracy deteriorates for regions with similar pixel values to other parts and large distance changes inside
Solution Approach 1:
The patent adds a new analysis dimension by examining spatial change characteristics of both pixel values and distances simultaneously. This multi-dimensional analysis (pixel value pattern + distance pattern) enables the system to distinguish target regions from background regions even when their absolute pixel values are similar, by analyzing how these values change across space.
Solution Approach 2:
The patent combines multiple analysis dimensions (pixel value spatial change characteristics and distance spatial change characteristics) into a composite extraction criterion. This composite approach leverages the complementary information from both pixel value patterns and distance patterns to achieve accurate extraction of regions with complex characteristics.
Data Source
AI summary
An image region extraction device for extracting a target region on the basis of a designated region with high precision. An image region extraction device (100) extracts a target region from an image on the basis of a designated region and is provided with: a spatial change learning unit (180) for acquiring, for each section of the image, a pixel value changing pattern which is a characteristic of the spatial changes in a pixel value component, and a distance changing pattern which is a characteristic of the spatial changes in a distance component extending from a reference position to a photographic object; and a region dividing unit (210) for extracting a target region on the basis of the similarities between the pixel value changing pattern and the distance changing pattern among the designated region and each section of the image.