Human Figure Region Extraction Using Face Position and Hierarchical Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting human figure regions from images are inaccurate due to inappropriate segmentation, where regions may be too small or large, leading to inclusion of background regions and poor judgment.
Innovation Solution
A method that detects a face or facial part in an image, determines a candidate region based on position information, judges unit regions to represent the human figure, and iteratively updates the estimated region to exclude background, using classifiers and skin color information for accurate extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If region segmentation processing is applied to the whole image, then human figure regions can be extracted automatically, but the accuracy of extraction deteriorates when segmentation degree is not appropriate
Solution Approach 1:
The patent divides the image processing into two stages: first dividing the image into multiple regions based on color distribution, then further dividing selected regions into unit regions for detailed judgment. This hierarchical segmentation allows automatic processing while maintaining accuracy by adapting the segmentation granularity to different regions.
Solution Approach 2:
The patent applies different processing strategies to different regions: regions with color similar to skin tone are divided into unit regions for detailed judgment, while other regions are handled differently. This local differentiation ensures that automatic extraction maintains high accuracy by focusing computational effort where it is most needed.
2Measurement precision
If regions are divided into small units, then accurate judgment of human figure portions is possible, but the number of regions increases and processing complexity increases
Solution Approach 1:
The patent implements a two-level segmentation system: first dividing the image into regions based on color distribution, then dividing only selected regions into unit regions. This reduces the total number of units that need detailed processing while maintaining judgment accuracy where it matters most.
Solution Approach 2:
The patent applies fine-grained unit region division only to areas with skin-like colors where accurate judgment is critical, while using coarser region-level processing elsewhere. This localized approach maintains judgment accuracy for human figure portions while reducing overall processing complexity.
3Device complexity
If regions are divided into large units, then processing complexity is reduced, but background regions may be included in the extracted human figure regions
Solution Approach 1:
The patent uses a hierarchical segmentation approach where the image is first divided into regions, then selected regions are further divided into unit regions. This ensures that even if initial regions are large, the subsequent unit region division eliminates background contamination while maintaining manageable processing complexity.
Solution Approach 2:
The patent applies detailed unit region judgment specifically to regions with skin-like colors, ensuring that background regions are not incorrectly included in human figure extraction. This localized detailed processing maintains extraction reliability without requiring fine-grained division of the entire image.
Data Source
AI summary
Upon extraction of a human figure region in an image, a face or facial part is detected in the image, and a candidate region that is deemed to include the human figure region is determined from position information of the detected face or facial part. Judgment is made as to whether each unit region having 2 pixels or more and comprising the candidate region represents the human figure region, and a set of the unit regions having been judged to represent the human figure region is determined as an estimated region which is estimated to include the human figure region. The human figure region is then extracted in the determined estimated region.