3D View Synthesis Dealiasing via Weighted Border Blending
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 3D view synthesis methods introduce aliasing in object borders due to the limitation of a single disparity value per pixel in disparity maps, leading to inaccurate representation of foreground and background information at object borders, resulting in aliased and unnatural transitions in rendered views.
Innovation Solution
A method and device that detect aliased borders in a disparity map, calculate a weighting/dealiasing coefficient for each pixel, and combine foreground and background video information based on this coefficient to smooth object borders, effectively replacing stair-like aliased contours with virtual straight lines, allowing for a two-layer representation of disparity values per border pixel.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single disparity value per pixel is used in the disparity map, then the data structure remains simple and storage is efficient, but aliasing appears at object borders and the representation of foreground and background information becomes inaccurate
Solution Approach 1:
The pixel is segmented into multiple regions (first region for foreground, second region for background) with different disparity values. This allows the disparity map to represent both foreground and background disparity information at object borders, eliminating aliasing while maintaining a manageable data structure through region-based segmentation rather than full sub-pixel resolution.
Solution Approach 2:
The invention transitions from representing disparity as a single scalar value per pixel to representing it as a function of spatial position within the pixel (x, y coordinates). This dimensional change allows the disparity map to capture continuous disparity variations across the pixel area, enabling accurate representation of object borders without increasing storage complexity through multiple discrete values.
2Measurement precision
If complex matting techniques are used to refine stereo depths at boundaries, then sub-pixel accuracy can be achieved, but the method becomes quite complex and relies on complex matting techniques
Solution Approach 1:
The invention performs preliminary segmentation of the pixel into foreground and background regions before depth refinement. By pre-defining the spatial distribution of foreground and background areas within each pixel, the subsequent depth refinement process becomes simpler and more direct, avoiding the need for complex iterative matting techniques while still achieving sub-pixel accuracy.
Solution Approach 2:
The invention changes the parameter representation from discrete disparity values to continuous disparity functions over pixel coordinates. This parameter transformation enables sub-pixel accuracy by allowing disparity to vary continuously across the pixel area, achieving precise boundary representation without relying on complex matting algorithms.
3Ease of manufacture
If object borders are located only at pixel borders in the disparity map, then the data structure remains simple, but aliasing appears and object borders exhibit stair-line artifacts
Solution Approach 1:
The invention moves object border representation from the discrete pixel border level to a continuous spatial level within pixels. By allowing disparity to vary with x, y coordinates within the pixel, the method achieves smooth object borders without stair-line artifacts while keeping the underlying data structure relatively simple through the use of separable horizontal and vertical components.
Data Source
AI summary
The present disclosure is directed to a method and a device for dealiasing borders in a 3D interpolated view, the interpolated view comprising at least foreground pixels with foreground video in formation and background pixels with background video information. The method comprises the step of detecting, in an interpolated disparity map, associated to the 3D interpolated view, an aliased border between the foreground pixels with a foreground disparity value and the background pixels with a background disparity value, the aliased border delimiting an inner aliased contour formed by foreground pixels and an outer aliased contour formed by background pixel, the aliased border having the form of a stair line; determining a weighting/dealiasing coefficient for each pixel of the outer contour said coefficient corresponding to the part of the pixel area included in the virtual right triangle defined by the two sides of each stair line to the area of the pixel; calculating video information for each pixel of the outer contour in combining the background video information and the foreground video information in a proportion defined by the determined weighting/dealiasing coefficient and replacing video information for each pixel of the outer contour by the calculated video information.