Spatio-Spectral Feature Learning for Shadow-Object Distinction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer systems face challenges in accurately distinguishing between shadows and material objects in images, which is crucial for applications like computer vision and robotics, as they lack the cognitive ability to automatically differentiate between the two based on pixel values alone.
Innovation Solution
A computerized method using spatio-spectral features, specifically the X-junctions where material edges and illumination boundaries intersect, is employed to identify material edges by calculating spectral ratios and applying machine learning techniques to classify these features, allowing for the differentiation between shadows and objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional pixel value analysis is used, then the processing is simple, but the ability to distinguish shadows from material objects is poor
Solution Approach 1:
The patent transitions from analyzing only spatial pixel values to incorporating spectral dimensions by capturing images across multiple wavelengths. This dimensional expansion allows the system to distinguish shadows from material objects by analyzing how different materials reflect light at various spectral wavelengths, thereby improving measurement precision without excessive complexity increase.
Solution Approach 2:
The patent changes the parameters being analyzed from simple intensity values to spectral ratios and spatio-spectral features. By computing ratios of pixel values across different wavelengths and analyzing these spectral characteristics, the system achieves better shadow-material distinction while managing complexity through focused feature extraction.
2Measurement precision
If spectral analysis is performed to improve shadow detection, then the accuracy improves, but the computational complexity increases
Solution Approach 1:
The patent extracts specific spectral features (ratios of pixel values at different wavelengths) from the full spectral data. By focusing computation on these key extracted features rather than processing all spectral information, the system achieves improved shadow detection accuracy while reducing the overall computational power required.
Solution Approach 2:
The patent performs preliminary spectral ratio calculations and feature extraction before the main shadow detection algorithm. By pre-computing spectral characteristics and organizing them into meaningful features, the system reduces the computational burden of the subsequent shadow-material distinction process.
3Loss of information
If multiple wavelength images are captured, then the spectral information improves, but the data processing time increases
Solution Approach 1:
The patent extracts essential spectral information by computing ratios between wavelengths rather than processing all spectral data in full detail. This extraction approach preserves the critical spectral information needed for shadow detection while significantly reducing the processing time compared to analyzing complete spectral datasets.
Solution Approach 2:
The patent transforms the raw spectral data from multiple wavelengths into spectral ratio parameters. This parameter transformation condenses the information from multiple wavelength images into compact spectral features that retain the essential spectral information while enabling faster processing.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
In a first exemplary embodiment of the present invention, an automated, computerized method is provided for determining illumination flux in an image. According to a feature of the present invention, the method comprises the steps of performing a computer learning technique to determine spatio-spectral information for the images, and utilizing the spatio-spectral information to identify illumination flux.