Depth Map Occultation Zone Filling via Directional Pixel Traversal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for filling occultation zones in depth or disparity maps often result in errors, artefacts, and loss of details, particularly around object contours, due to the partial addressing of defects and inconsistencies in reconstructed images.
Innovation Solution
A method involving pixel-by-pixel traversal of the map, where invalid pixels are assigned values based on the highest depth or lowest disparity from the first valid pixel following them, ensuring background propagation and preventing object details from being propagated incorrectly, especially in concave objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing techniques are used to fill occultation zones in depth or disparity maps, then the zones can be filled, but errors and artefacts are introduced, particularly around object contours
Solution Approach 1:
The patent segments the filling process into distinct phases: identifying invalid pixels, searching for valid reference pixels in specific directions, and applying different filling strategies based on the pixel's location and context. This segmentation allows precise control over what information is propagated and what is preserved, eliminating errors around object contours while filling occultation zones
Solution Approach 2:
The patent applies local quality by treating different regions of the map differently based on their characteristics. Valid pixels are preserved with their original values, while invalid pixels are filled using context-appropriate strategies. The directional search and neighborhood analysis ensure that local structures and object contours are respected in the filling process
2Productivity
If simple filling methods are applied to occultation zones, then processing is fast, but object details are lost or incorrectly propagated
Solution Approach 1:
The patent performs preliminary actions by first identifying all invalid pixels and their locations before filling them. The directional search for valid reference pixels is prepared in advance, and the filling strategy is determined based on pre-analyzed neighborhood characteristics. This preliminary organization enables efficient processing while maintaining detail precision
Solution Approach 2:
The patent introduces directional searching in multiple dimensions (horizontal, vertical, diagonal directions) to find valid reference pixels. This multi-dimensional approach allows the algorithm to explore more potential source pixels without significantly increasing computational complexity, thereby preserving object details while maintaining processing efficiency
3Ease of operation
If background propagation is performed without directionality, then filling is simple, but object contours become distorted
Solution Approach 1:
The patent applies asymmetry by implementing directional search patterns that are sensitive to the geometry of occultation zones. The search directions are asymmetrically chosen based on the location and context of invalid pixels, allowing the algorithm to respect object contours while filling background areas. This asymmetric approach maintains contour accuracy without sacrificing operational simplicity
Data Source
Figure 1~2b
Figure 3~4b
Figure 5a
AI summary
A method of filling in the zones of occultation of a map of depth or of disparities, the map being estimated on the basis of at least two images and being made up of a matrix of pixels forming a set of rows and columns, each pixel of the map being associated with a depth or disparities value, termed the "pixel value", and any pixel of an occultation zone, termed an "invalid pixel", being associated with a value of a pixel identifiable as being invalid. This method comprises the pixel-by-pixel traversal (E1, E2, E10) of each row of the map in a predefined direction, and the assigning to each invalid pixel Pi encountered (E3) on a current row, of a value determined (E4-E9) as a function of the values associated with the pixels belonging to a predefined neighbourhood around the first valid pixel P1, if it exists, which follows the invalid pixel Pi on the current row and whose value, V(P1), corresponds to a greater depth or to a lower disparity with respect to the value, V(PO), of the last valid pixel, PO, if it exists, which precedes the invalid pixel Pi on the current row.