Fractional-Pixel Image Shifting for Accurate Disparity Calculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing block matching methods for calculating disparity are inaccurate due to luminance value deviations between images, leading to incorrect disparity calculations even after filter processing.
Innovation Solution
An image processing apparatus that generates shifted standard and reference images in fractional pixel units, calculates correlation values between these shifted images, and determines disparity based on these values without fitting to a function, thereby reducing errors caused by luminance deviations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Illumination intensity
If filter processing is applied to correct luminance value deviations, then luminance correction is improved, but disparity calculation accuracy deteriorates due to incorrect first disparity calculation
Solution Approach 1:
The patent applies filter processing to the A-image and B-image before performing the first disparity calculation. This preliminary filtering corrects luminance value deviations in advance, ensuring that the subsequent disparity calculation uses corrected images, thereby preventing propagation of luminance-related errors through the processing pipeline.
Solution Approach 2:
The patent divides the disparity calculation into multiple stages: first disparity calculation using filtered images, second disparity calculation using unfiltered images, and iterative refinement. This segmentation allows different processing strategies to be applied at different stages, improving overall accuracy while managing computational complexity.
2Measurement precision
If subpixel estimation method is used to calculate disparity in fractional pixel units, then disparity precision is improved, but calculation complexity increases due to function fitting requirements
Solution Approach 1:
The patent replaces the traditional function-fitting approach of subpixel estimation with a correlation-based method. Instead of fitting correlation values to a function to estimate subpixel disparity, the patent calculates correlation values between shifted images at fractional pixel positions directly, simplifying the computational process while maintaining subpixel precision.
Solution Approach 2:
The patent changes the approach from function fitting to direct correlation calculation at fractional pixel positions. By evaluating correlation at specific fractional shifts rather than fitting a continuous function, the method achieves subpixel accuracy with reduced computational complexity and avoids issues with luminance deviation affecting function fitting.
3Stability of the object's composition
If luminance value deviation correction is performed, then image consistency is improved, but incorrect disparity results occur when deviations from correction values are present
Solution Approach 1:
Filter processing is applied to correct luminance value deviations before disparity calculation. This preliminary correction improves image consistency and reduces luminance-related errors in the subsequent correlation-based disparity estimation.
Solution Approach 2:
The patent performs disparity calculation in multiple stages with different processing approaches. The first disparity calculation uses filtered images for robustness, while the second uses unfiltered images for precision, and iteratively refines the result to compensate for any correction deviations.
Data Source
AI summary
An image processing apparatus configured to calculate a disparity value from first and second images includes at least one processor or circuit configured to function as an image setting unit configured to set a specific region in the first image as a standard image and a specific region in the second image as a reference image, a shifted image generating unit configured to generate a shifted standard image by shifting the standard image based on a first shift amount and generate a shifted reference image by shifting the reference image based on a second shift amount, and a disparity calculating unit configured to calculate the disparity value based on the first and second shift amounts. The shifted image generating unit sets one of the first and second shift amounts in fractional pixel units.