Face Image Conversion for Accurate Skin Condition Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing skin condition estimation models are hindered by the inclusion of non-skin regions in face images, leading to decreased estimation accuracy due to the need to learn irrelevant information and reduced resolution of the face region during normalization.
Innovation Solution
A method that converts face images to a predetermined rectangular shape, positioning face region contours at the image edge, excluding non-skin regions like eyes and lips, to improve estimation accuracy by maintaining higher resolution and focusing solely on skin areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the face image is normalized to a predetermined size including the entire face region, then the image can be processed by the estimation model, but the resolution of the face region decreases and non-skin regions are included
Solution Approach 1:
The patent extracts only the necessary skin region from the face image by positioning contour points at the image edge, excluding non-skin regions such as eyes, lips, and background. This extraction principle resolves the contradiction by removing irrelevant areas that would otherwise reduce estimation accuracy while maintaining adequate resolution of the remaining skin region.
Solution Approach 2:
The patent transforms the image representation by converting contour coordinates to edge-positioned coordinates, effectively changing the dimensional mapping of the face region. This dimensionality change allows the same amount of image data to represent the skin region more efficiently, maintaining resolution while enabling accurate estimation.
2Area of stationary object
If the face image includes non-skin regions such as eyes, lips, and background, then the complete face area is captured, but the estimation model must learn irrelevant information reducing accuracy
Solution Approach 1:
The patent applies the extraction principle by removing non-skin regions from the image through coordinate transformation. By positioning contour points at the image edge, the transformation naturally excludes areas such as eyes, lips, and background, allowing the model to focus solely on relevant skin information and improve estimation accuracy.
Solution Approach 2:
The patent applies local quality by giving different treatment to different regions of the face. The skin region is preserved and enhanced with proper resolution, while non-skin regions are excluded. This selective regional processing ensures that only relevant areas contribute to the estimation, resolving the contradiction between capturing complete face area and maintaining estimation accuracy.
3Adaptability or versatility
If the image is normalized to a fixed size, then the input rule of the estimation model is satisfied, but the face region resolution is reduced
Solution Approach 1:
The patent resolves this contradiction by changing the coordinate dimensionality of the face region. Instead of simply resizing the entire image, the contour points are transformed to edge positions, creating a new coordinate system that maintains resolution while satisfying the fixed-size input requirement of the estimation model.
Solution Approach 2:
The patent applies parameter changes by modifying the coordinate parameters of the face region through mathematical transformation. The contour coordinates are changed to edge-positioned coordinates, which allows the image to maintain higher resolution while still conforming to the predetermined size requirement for model input.
Data Source
AI summary
In a skin condition estimating method, one or more processors, capable of using a learned model for estimating a skin condition of a human face appearing in an input face image from the input face image having a predetermined rectangular shape with a predetermined image size, execute a process of acquiring a face image of a subject, a process of identifying position coordinates of a plurality of points in a face region of a human face appearing in the acquired face image, a process of converting the face image by coordinate-converting some or all of the position coordinates of the identified plurality of points so that the face region has a predetermined image size and a predetermined rectangular shape, and a process of inputting the converted face image to the learned model to acquire skin condition information of the face of the subject.