Multiview Image Hole Filling Using Depth-Based Patch Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multiview image generating methods distort images, especially when the depth of the background around the hole area is high, due to simple filling techniques that do not effectively process structural portions of the original image.
Innovation Solution
A method that generates a hole image using depth information, divides the boundary region into patches, selects areas with similar pixel values to the background, and fills the hole areas with corresponding pixel values to create a smooth multiview image, using techniques such as scaling and comparing pixel values, colors, depths, and gradients to determine the most similar areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If simple filling techniques are used to generate multiview images, then the processing speed is fast and the method is simple, but the image quality deteriorates with distortion especially in high depth background areas
Solution Approach 1:
The patent divides the image into multiple layers based on depth information, separating foreground and background regions. This segmentation allows different filling strategies to be applied to different depth regions, improving image quality in high-depth areas while maintaining processing efficiency through selective processing.
Solution Approach 2:
The patent applies different filling techniques to different regions of the image based on local depth characteristics. High-depth background areas receive more sophisticated processing (e.g., texture synthesis, gradient-based filling) while low-depth areas use simpler methods, optimizing both quality and speed locally.
2Manufacturing precision
If sophisticated filling techniques are used to reduce distortion, then the image quality improves, but the processing complexity and time increase
Solution Approach 1:
The patent performs preliminary depth map generation and region classification before the actual filling process. By pre-identifying high-depth background areas and preparing appropriate filling patterns in advance, the system reduces the complexity of the main filling operation while maintaining high image quality.
Solution Approach 2:
The patent adjusts filling parameters dynamically based on depth information. For example, the search radius for similar regions, the number of iterations for optimization, and the blending coefficients are modified according to the local depth value, allowing sophisticated processing only where necessary.
3Measurement precision
If depth information is utilized for hole filling, then the accuracy of structural portions improves, but the computational requirements increase
Solution Approach 1:
The patent applies accurate depth-based filling only to necessary regions (hole areas and their boundaries) rather than processing the entire image. By limiting sophisticated processing to localized regions where it is most needed, the computational energy consumption is reduced while maintaining high filling accuracy.
Solution Approach 2:
The patent uses texture copying and patch matching techniques where regions from the original image are copied to fill holes. Depth information guides the selection of source regions, ensuring structural accuracy while the copying operation itself is computationally efficient compared to generative methods.
Data Source
AI summary
A multiview image generating method is provided. The multiview image generating method includes generating a hole image using depth information of an original image, dividing a boundary region in which a hole area of the generated hole image is in contact with a background into at least one patch, selecting an area having the most similar pixel value to a background included in the divided at least one patch, in a preset area of the original image, and filling a hole area on the divided at least one patch with a pixel value of the selected area to generate a first view image.