Image Evaluation Device for Multi-Face Subjective Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image evaluation methods fail to accurately reflect subjective viewer evaluations and do not effectively handle images containing multiple faces, as they rely on quantitative calculations rather than user-centric assessments.
Innovation Solution
An image evaluation device and method that acquires and statistically calculates individual evaluation values for various face-related information such as number, size, position, orientation, and detection score, using evaluation tables based on user-selected samples to provide a more subjective and accurate evaluation, including positional relationships and front face ratios for multiple faces.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image evaluation is performed using quantitative calculation methods (evaluation values for lightness, face ratio, orientation, etc.), then the evaluation process is objective and systematic, but the results do not reflect subjective evaluation by actual viewers
Solution Approach 1:
The patent implements feedback by collecting actual viewer evaluation results for training images and using this feedback to train the neural network. The system continuously improves its evaluation accuracy by incorporating real human judgment data, allowing the quantitative evaluation to better reflect subjective viewer preferences while maintaining systematic processing
Solution Approach 2:
The patent replaces traditional mechanical calculation methods (summing evaluation values for lightness, face ratio, orientation) with a neural network-based system. This substitution allows the system to learn complex patterns from training data and produce evaluation results that better match subjective viewer preferences without sacrificing the objectivity and systematic nature of automated evaluation
2Productivity
If simple evaluation value calculation is used for images with multiple faces, then the processing is simple and fast, but appropriate evaluation of images containing more than one face is not provided
Solution Approach 1:
The patent applies dynamics by making the evaluation process adaptive to the number of faces in the image. The neural network dynamically adjusts its evaluation based on the detected number of faces, using different evaluation criteria and weighting for single-face versus multi-face images. This allows the system to maintain processing speed while improving evaluation accuracy for multi-face images through context-aware adaptation
3Measurement precision
If comprehensive evaluation of multiple face attributes (number, size, position, orientation, rotational angle, detection score) is performed, then the evaluation is thorough and detailed, but the complexity of the evaluation system increases
Solution Approach 1:
The patent segments the evaluation process into distinct functional modules: face detection module, attribute extraction module (for number, size, position, orientation, rotational angle, detection score), neural network evaluation module, and result output module. This segmentation allows comprehensive evaluation of multiple attributes while managing system complexity through modular design, where each module handles a specific aspect of the evaluation independently
Data Source
AI summary
An image evaluation device includes: an information acquiring unit to acquire, from an image containing at least one face, at least one type of information including at least the number of the at least one face and optionally including any of a size of the face, a position of the face in the image, an orientation of the face, a rotational angle of the face and a detection score of the face; and an individual evaluation value calculating unit to statistically calculate an individual evaluation value indicating a result of evaluation for each type of information based on the acquired information.