2D to Stereoscopic Image Conversion Using Face Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing two-dimensional to stereoscopic conversion processes face challenges in precisely analyzing and separating foreground objects from the background, leading to incorrect depth assignments and a less than satisfactory stereoscopic experience due to difficulties in k-means segmentation.
Innovation Solution
A method that detects faces and body regions in images, generates color or feature models, calculates similarity values of image pixels based on these models, and assigns depth values to generate stereoscopic images, improving the accuracy of depth placement for foreground objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If k-means segmentation is used for separating foreground objects from background, then the conversion process can be automated, but the segmentation precision deteriorates leading to incorrect depth assignments
Solution Approach 1:
The patent segments the image processing task into multiple specialized stages: face detection, body region determination, color/feature model generation, similarity calculation, and depth assignment. Each stage handles a specific aspect of the foreground-background separation problem, improving overall segmentation precision while maintaining automation.
Solution Approach 2:
The patent introduces color models and feature models as intermediary representations between the raw image data and depth assignments. These intermediaries capture the essential characteristics of foreground objects (particularly human figures) and enable more accurate separation from the background through similarity comparison.
2Productivity
If simple depth assignment algorithms are used, then the processing speed is maintained, but the depth placement accuracy deteriorates resulting in a less than satisfactory stereoscopic experience
Solution Approach 1:
The patent performs preliminary actions by detecting faces and determining body regions before depth assignment. This pre-processing identifies foreground objects with high confidence, allowing subsequent depth algorithms to focus computational resources on these identified regions and achieve better accuracy without significantly increasing overall processing time.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image. Face regions and body regions receive specialized color/feature model-based processing, while other regions may use simpler methods. This local differentiation improves depth placement accuracy for critical foreground elements while maintaining processing efficiency.
Data Source
AI summary
In one embodiment, a method of two-dimensional to stereoscopic image conversion, the method comprising detecting a face in a two-dimensional image; determining a body region based on the detected face; providing a color model from a portion of the determined body region, a portion of the detected face, or a combination of both portions; calculating a similarity value of at least one image pixel of the two-dimensional image based on the provided color model; and assigning a depth value of the image pixel based on the calculated similarity value to generate a stereoscopic image.