Image Token Analysis for Material and Shadow Boundary Distinction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional algorithms for image processing struggle to accurately distinguish between shadow and material object edges, leading to false positives and false negatives due to the assumption that shadow boundaries are soft and material object edges are sharp, which is not always the case in real-world scenarios.
Innovation Solution
The method involves identifying Type C tokens in an image, which are connected regions of pixels with similar characteristics, and then constructing Type B tokens by analyzing and merging these tokens using techniques like arbitrary boundary removal, adjacent planar token merging, and local token analysis to accurately represent regions of single materials under varying illumination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional algorithms assume sharp boundaries indicate material edges and soft boundaries indicate shadows, then shadow and object edge recognition can be implemented, but false positives and false negatives occur when shadows form sharp boundaries or material edges form soft boundaries
Solution Approach 1:
The patent transforms the image from RGB color space to a different color space (such as Lab or XYZ) and applies spectral ratio analysis. By changing the parameter representation of pixel values and analyzing the ratio of spectral components, the algorithm can distinguish material edges from shadow boundaries regardless of boundary sharpness, resolving the contradiction between recognition accuracy and false positive rate
Solution Approach 2:
The patent replaces the mechanical thresholding approach (comparing boundary sharpness against fixed criteria) with a spectral analysis mechanism. Instead of relying on the mechanical assumption that shadows are always soft and edges are always sharp, the system uses spectral ratio calculations to identify material boundaries, eliminating false positives and false negatives caused by boundary characteristics
2Measurement precision
If spectral ratio analysis is used to distinguish material boundaries from illumination boundaries, then accuracy improves, but computational complexity increases due to deriving characteristic spectral ratios from tokens
Solution Approach 1:
The patent segments the image into tokens (connected regions of similar pixels) before performing spectral ratio analysis. By dividing the image into meaningful segments first, the algorithm reduces the computational scope of spectral analysis to only boundary regions, maintaining high accuracy while reducing overall computational complexity compared to analyzing every pixel in the entire image
3Adaptability or versatility
If region growing and merging algorithms are used for color segmentation, then spatially disconnected but colorimetrically similar regions can be identified, but the segmentation may not correspond to actual material boundaries under varying illumination
Solution Approach 1:
The patent introduces spectral ratio analysis as an intermediary step between region segmentation and material boundary identification. Instead of directly merging regions based on color similarity, the system first calculates spectral ratios at boundaries to determine whether they represent material edges or illumination variations. This intermediary analysis ensures that region merging corresponds to actual material boundaries even under varying illumination conditions
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
In a first exemplary embodiment of the present invention, an automated, computerized method is provided for processing an image. According to a feature of the present invention, the method comprises the steps of identifying token regions in the image, each token region comprising a connected image region of similar color and intensity, and utilizing the identified token regions to generate single material token regions for use in processing material and illumination aspects of the image.