Linear Token Identification for Shadow Edge Differentiation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for distinguishing between shadows and material object edges in images often result in false positives and false negatives due to the inability to accurately differentiate between sharp shadow boundaries and material object edges, as shadows can form sharp boundaries and material edges can be soft.

Innovation Solution

The method involves identifying linear tokens in an image, calculating characteristic spectral ratios, and using illumination gradients to differentiate between shadow and material edges, employing a bi-illuminant dichromatic reflection model to classify pixel clusters and identify illumination fields, which helps in accurately distinguishing between shadows and material edges.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If conventional brightness boundary analysis is used to detect edges, then the method can be implemented by algorithms that can be accurately executed by a computer, but the results will often be incorrect due to inability to distinguish between sharp shadow boundaries and material object edges

Engineering Contradiction:
Improvealgorithm executabilityVSAvoidedge detection accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent changes the parameters used for edge detection from simple brightness boundaries to spectral ratios and illumination gradients. By analyzing the spectral characteristics (ratios of different wavelength bands) and gradient directions, the system can distinguish between shadow boundaries (which have specific spectral ratio patterns) and material edges (which have different patterns), thereby improving detection accuracy while remaining computationally executable

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent introduces spectral ratio analysis as an intermediary step between raw brightness data and edge detection results. By computing ratios of pixel intensities across different spectral bands and comparing them against expected patterns for shadows versus material edges, the system creates an intermediate representation that enables more accurate discrimination before final edge identification

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If spectral ratio analysis is used to identify illumination boundaries, then accurate differentiation between shadows and material edges is achieved, but the computational complexity and processing requirements increase

Engineering Contradiction:
Improveshadow and edge differentiation accuracyVSAvoidprocessing system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the image processing task into distinct stages: first identifying candidate regions using simplified criteria, then applying full spectral ratio analysis only to those regions, and finally classifying them based on gradient and spectral patterns. This segmentation allows complex analysis to be applied selectively rather than uniformly across the entire image, reducing overall computational complexity while maintaining high accuracy where needed

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP2752791B8Method for identifying linear tokens in an image
Publication Date: 2019.10.30 TANDENT COMPUTER VISION LLC

AI summary

A method for identifying linear tokens in an image is disclosed, a linear token being a connected region in the image wherein adjacent pixels of the region have differing color measurement values that fall within a cylinder in a color space with a positive slope from a dark end to a bright end for each color channel of the image. For each pixel, a local area around the pixel is tested to determine whether the color values of the pixel and of its neighboring pixels fall within a linear segment or cylinder in the color space. If this is the case, the set of pixels is designated a linear token. The identification of linear tokens can be improved by optional additional filtering steps. The cylinder in color space is predicted by a bi-illuminant dichromatic reflection model, taking into account an ambient illuminant in addition to a direct illuminant.