Pixel Neighborhood Dissimilarity for Stereo Vision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing pixel dissimilarity measures are insensitive to noise, sampling artifacts, and depth discontinuities, leading to imprecise comparisons in image analysis tasks.
Innovation Solution
Computing multidimensional dissimilarity measures based on pixel neighborhood values from spatially-shifted neighborhoods in both images, using interpolation to preserve local smoothness and reduce noise, and selecting the smallest differences between extrema as the dissimilarity metric.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional pixel dissimilarity measures are used, then computation is simple and fast, but measurement precision deteriorates due to sensitivity to noise, sampling artifacts, and depth discontinuities
Solution Approach 1:
The patent divides the pixel comparison into multiple segments by computing dissimilarity measures at multiple sub-pixel positions along the epipolar line. Instead of comparing single pixels directly, the method segments the comparison into multiple points where intensity functions are evaluated and interpolated, thereby reducing sensitivity to noise and sampling artifacts while maintaining computational feasibility through structured processing
Solution Approach 2:
The patent transitions from one-dimensional pixel comparison to two-dimensional evaluation by introducing the epipolar line dimension. The method compares intensity functions along the epipolar line connecting corresponding pixels in stereo images, adding a spatial dimension that provides additional information about depth and geometric relationships, thereby improving measurement precision without excessive computational cost
2Reliability
If pixel dissimilarity measures insensitive to gain, bias, noise, and depth discontinuities are used, then robustness improves, but measurement precision deteriorates due to insensitivity to sampling variations
Solution Approach 1:
The patent changes the parameters of the dissimilarity measurement by using linearly interpolated intensity functions instead of direct pixel values. The method evaluates intensity functions at multiple points along the epipolar line and uses interpolation to estimate values at sub-pixel positions, thereby making the measurement sensitive to sampling variations while remaining robust to noise and depth discontinuities through the smoothing effect of linear interpolation
Data Source
AI summary
Methods, machines, and machine-readable media for computing dissimilarity measures are described. In one aspect, a first set of pixel neighborhood values (PNVs) is computed from respective sets of pixel values of a first image corresponding to different spatially-shifted pixel neighborhoods each encompassing a mutual target pixel in the first image. A second set of PNVs is computed from respective sets of pixel values of a second image corresponding to different spatially-shifted pixel neighborhoods each encompassing a mutual target pixel in the second image. A measure of dissimilarity is computed based at least in part on the first and second sets of computed PNVs. The computed dissimilarity measure is stored on a machine-readable medium.