2D to 3D Conversion via Scene Classification and Face Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 2D to 3D conversion techniques rely solely on low-level features for depth map generation, resulting in inaccurate depth maps and uncomfortable viewing experiences due to artifacts like geometric inconsistencies and shearing in the rendered 3D images.
Innovation Solution
The method incorporates high-level image features such as face detection and scene classification to enhance depth maps, using a Depth Based Image Rendering algorithm to assign accurate depth values to regions in 2D images, thereby improving object segmentation and reducing inconsistencies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If only low-level features are used for depth map generation, then the processing complexity is reduced, but the depth map accuracy deteriorates causing geometric inconsistencies and shearing artifacts
Solution Approach 1:
The image is segmented into multiple regions based on scene classification (sky, foliage, ground, water, snow, ice, building, vehicle, person, animal). Each region is then processed independently with appropriate depth estimation techniques, allowing high accuracy without processing the entire image uniformly, thus reducing overall complexity
Solution Approach 2:
Different depth estimation methods are applied to different regions based on their characteristics. For example, face detection is applied to person regions while other regions use standard depth estimation. This localized approach improves accuracy for specific regions without requiring the most complex methods for the entire image
2Measurement precision
If high-level features like face detection and scene classification are incorporated, then depth map accuracy is improved, but processing time increases
Solution Approach 1:
Scene classification is performed as a preliminary step to divide the image into regions before applying depth estimation. This preliminary segmentation allows subsequent depth processing to be optimized for each region, reducing the time required for complex operations on the entire image
Solution Approach 2:
Face detection is applied selectively only to regions identified as containing persons, rather than processing the entire image. This partial application of computationally intensive face detection algorithms reduces overall processing time while maintaining high accuracy for relevant regions
Data Source
AI summary
A method and apparatus for converting a two dimensional image or video to three dimensional image or video. The method includes segmenting objects in at least one of a two dimensional image and video, performing, in the digital processor, a depth map generation based on low-level features, performing face detection and scene classification on at least one of a two dimensional image and video, and utilizing face detection and scene classification in enhancing the depth map and for converting the at least one of a two dimensional image and video to three dimensional image and video.