Guided Interpolation for Depth Map Boundary Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for generating depth information maps, such as disparity maps, face challenges in preserving image details and object boundaries, often resulting in blurred boundaries and increased computational complexity, particularly with double Gaussian filters and CID approaches.
Innovation Solution
A method and apparatus for generating a final depth information map through guided interpolation, which involves receiving a coarse depth map with lower resolution, performing adaptive low-pass filtering and decimation to reduce noise, and using a guided interpolation block to reconstruct the map with enhanced object boundary alignment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional CID approach with block-based operations is used to generate depth map, then computational complexity is reduced, but image details are lost and object boundaries become blurred
Solution Approach 1:
The patent segments the depth map generation process into two distinct stages: coarse depth map generation using block-based CID operations, and subsequent refinement using guided filtering. This segmentation allows each stage to specialize - the first for computational efficiency and the second for boundary precision, thereby resolving the contradiction between complexity and precision.
Solution Approach 2:
The patent performs preliminary coarse depth map generation using computationally simple block-based operations before applying the more complex guided filtering refinement. This preliminary action establishes a foundation that reduces subsequent processing complexity while still allowing for detailed boundary refinement later.
2Manufacturing precision
If double Gaussian filters are used in bilateral filtering approach to preserve object boundaries, then object boundary precision is improved, but computational complexity and required computing power increase significantly
Solution Approach 1:
The patent extracts and isolates the guided filtering operation as a separate refinement stage applied only to the coarse depth map. This extraction avoids applying complex double Gaussian filters to the entire high-resolution image, thereby reducing computational complexity while maintaining boundary precision where it matters most.
Solution Approach 2:
The guided filtering approach applies different processing characteristics to different regions: it preserves sharp object boundaries where needed while applying smoother filtering in homogeneous regions. This local quality differentiation maintains boundary precision without requiring excessive computational resources across the entire image.
3Adaptability or versatility
If conventional guided filtering approach is used without linear kernel, then adaptability to different depth characteristics is improved, but burden of adjusting output image characteristics increases
Solution Approach 1:
The patent introduces a linear kernel parameter in the guided filtering process that controls the blurriness and output characteristics of the depth map. By changing this parameter, users can easily adjust the output characteristics to match different depth scenarios, thereby improving ease of operation while maintaining adaptability through the flexible parameter control.
Data Source
AI summary
A method for generating a final depth information related map includes the following steps: receiving a coarse depth information related map, wherein a resolution of the coarse depth information related map is smaller than a resolution of the final depth information related map; and outputting the final depth information related map reconstructed from the coarse depth information related map by receiving an input data and performing a guided interpolation operation upon the coarse depth information related map according to the input data.