Skin Selection via Bounding Box and Object Mask Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for selecting skin in digital images often mistakenly select non-skin portions with similar colors, leading to tedious manual corrections and a lack of visually pleasing digital content featuring exposed skin.
Innovation Solution
An image processing system determines bounding boxes for individuals in the image, generates object masks within these boxes, and identifies pixels with skin-like colors to selectively edit exposed skin without including other skin-colored areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional color-based skin selection is used, then skin-colored portions are selected, but non-skin portions with similar colors are also incorrectly selected
Solution Approach 1:
The patent segments the image processing into distinct stages: first identifying person locations using bounding boxes, then generating object masks for each person, and finally selecting skin-colored pixels only within those masks. This segmentation prevents non-skin portions from being incorrectly selected by confining the color-based selection to person-specific regions.
Solution Approach 2:
The patent introduces object masks as an intermediary layer between the person detection and skin selection processes. These masks act as a filter that allows only relevant pixels (those belonging to persons) to be considered for skin color selection, thereby eliminating false selections of non-skin portions with similar colors.
2Productivity
If conventional skin selection systems are used, then skin-colored areas are selected, but manual correction is required to remove non-skin portions
Solution Approach 1:
The patent performs preliminary actions by first detecting person locations and generating object masks before executing the skin color selection. This preliminary segmentation ensures that when skin-colored pixels are selected, they are automatically confined to correct regions, eliminating the need for subsequent manual correction of non-skin portions.
3Measurement precision
If bounding box-based person identification is implemented, then skin selection accuracy improves, but processing complexity increases
Solution Approach 1:
The patent divides the complex task of skin selection into manageable segments: bounding box detection, object mask generation, and skin color selection within masks. This segmentation makes the overall system more tractable and implementable while maintaining high accuracy.
Solution Approach 2:
The object mask generation component serves multiple functions: it defines person boundaries, confines skin color selection to relevant regions, and prevents selection of non-skin portions. This multi-functionality reduces the need for separate processing steps, thereby managing system complexity.
Data Source
AI summary
Depicted skin selection is described. An image processing system selects portions of a digital image that correspond to exposed skin of persons depicted in the digital image without selecting other portions. Initially, the image processing system determines a bounding box for each person depicted in the digital image. Based solely on the portion of the digital image within the bounding box, the image processing system generates an object mask indicative of the pixels of the digital image corresponding to a respective person. Portions of the digital image outside the bounding box are not used for generating this object mask. The image processing system then identifies the pixels of the digital image indicated by the object mask and having a similar color to a range of exposed skin colors determined for the respective person. The processing system generates skin selection data describing the identified pixels and enabling the exposed skin selection.