Face Image Attribute Conversion Identity Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques often impair the identity of the original image during attribute conversion, such as age conversion, leading to undesirable outcomes.
Innovation Solution
An image processing device and method that calculates similarity scores and distributions for multiple attribute conversions, acquiring conversion parameters to generate a new image while maintaining the original image's identity by comparing score distributions and using these parameters to convert the image attributes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If attribute conversion is performed on the original image using existing image processing techniques, then the attribute of the image is converted, but the identity of the original image is impaired
Solution Approach 1:
The patent employs feedback mechanisms by calculating similarity scores between the converted face images and reference face images, then using these scores to adjust and select the optimal conversion parameters. The similarity score calculation provides feedback on how well the converted image maintains identity, allowing the system to iteratively refine the conversion process to preserve identity while achieving attribute conversion.
Solution Approach 2:
The patent changes multiple parameters simultaneously including age, gender, ethnicity, and other attributes during face image conversion. By controlling and adjusting these parameters systematically and selecting conversion parameters that maintain similarity scores above a threshold, the system achieves versatile attribute conversion while preserving the original identity characteristics.
2Measurement precision
If multiple conversion methods are applied to convert image attributes, then the conversion accuracy is improved, but the calculation complexity increases
Solution Approach 1:
The patent calculates similarity scores for multiple conversion methods (excessive action) but only selects the optimal conversion parameters based on the highest similarity scores (partial action). This approach ensures high conversion accuracy by evaluating multiple methods while managing complexity by focusing computational resources on selecting rather than fully executing all conversion methods.
Solution Approach 2:
The patent segments the conversion process into distinct stages: calculating similarity scores for multiple conversion methods, acquiring score distributions, comparing distributions, and selecting optimal conversion parameters. This segmentation allows the system to handle multiple conversion methods systematically, improving accuracy through comprehensive evaluation while managing complexity through structured processing steps.
Data Source
AI summary
In the image processing device, the score calculation means calculate a first similarity score group indicating similarity between a first face image and each of n face images, and m second similarity score groups indicating similarity between m second face images obtained by converting an attribute of the first face image and each of the n face images. The score distribution acquisition means acquires the first score distribution indicating a distribution state of the first similarity score group, and the m second score distributions indicating a distribution state each of the m second similarity score groups. The conversion parameter acquisition means acquires one conversion parameter based on a comparison result between the first score distribution and the m second score distributions. The image generation means generates a face image by converting the attribute of the first face image based on the conversion parameter.