Confidence Coefficient Calculation for Depth Image Boundary Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for correcting errors in depth and motion information at object boundaries in images fail to accurately evaluate pixel reliability, leading to insufficient correction due to the lack of proper confidence coefficient calculation.
Innovation Solution
A data processing apparatus and method that calculates a confidence coefficient for pixels at object boundaries, using similarity and statistical analysis to differentiate between error-prone and error-free regions, allowing for accurate correction of depth image data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If template matching technique is used to determine depth information or motion information, then correspondence between two images can be acquired, but a major error is generated in the boundary of objects where one template includes objects of different depths and motions
Solution Approach 1:
The patent applies local quality by differentiating confidence coefficient calculation between object boundary regions and non-boundary regions. For boundary pixels, the confidence coefficient is calculated considering only pixels from the same object (same depth/motion), while for non-boundary pixels, all peripheral pixels are used. This local differentiation resolves the contradiction by ensuring high reliability in boundary regions without sacrificing measurement precision in non-boundary regions.
2Reliability
If confidence coefficient is calculated based on brightness value magnitude and change, then confidence can be reduced in areas with no texture or large motion, but confidence coefficient is calculated high in object boundaries with texture and small motion, failing to consider object boundaries
Solution Approach 1:
The patent segments the image processing into two distinct paths: boundary pixel processing and non-boundary pixel processing. The segmentation is achieved by detecting object boundaries and applying different confidence coefficient calculation methods to each segment. This resolves the contradiction by ensuring that boundary pixels receive specialized handling that prevents erroneous high confidence values, while non-boundary pixels continue to benefit from the brightness-based confidence calculation.
3Measurement precision
If clustering is performed and representative depth value is used for correction, then spatial continuity of pixel values can be considered, but if error is included in depth values in the class, the representative depth value is shifted from the correct depth value and correction becomes insufficient
Solution Approach 1:
The patent applies preliminary action by calculating confidence coefficients before performing the correction process. The confidence coefficient serves as a pre-evaluation metric that identifies reliable pixels and unreliable pixels (including boundary pixels with errors). This preliminary assessment allows the correction process to selectively apply representative depth values only to pixels with sufficient confidence, preventing erroneous corrections that would shift the representative depth value away from the correct value.
4Reliability
If weighted cross-bilateral filter is used with confidence coefficient, then depth image can be corrected using peripheral pixel information, but the confidence coefficient is simply a relative confidence coefficient among pixels, not the confidence coefficient of depth information itself
Solution Approach 1:
The patent applies parameter changes by modifying the confidence coefficient calculation to incorporate object boundary information and depth consistency checks. The confidence coefficient is transformed from a simple relative measure among pixels to a meaningful indicator of depth information reliability. This is achieved by comparing depth values of peripheral pixels with the target pixel and adjusting the confidence coefficient accordingly, ensuring that the coefficient reflects actual depth information quality rather than just relative pixel relationships.
Data Source
Figure 1
Figure 2A~2C
Figure 3
AI summary
Provided is a data processing apparatus for calculating confidence coefficient data to indicate a confidence of pixel values of a second image data. The apparatus acquires similarity between a pixel value of a first pixel of the first image data corresponding to a target pixel of the confidence coefficient data, and each pixel value of a plurality of second pixels around the first pixel. The apparatus acquires the confidence coefficient by determining an estimated pixel value in the third pixel of the second image data corresponding to the first pixel, based on the similarity of each of the plurality of second pixels, each pixel value of the plurality of fourth pixels around the third pixel and the pixel value of the third pixel and by comparing the pixel value of the third image and the estimated pixel value in the third pixel.