Cover Image Selection Using Aesthetic Scoring Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for obtaining cover images and training image scoring models in computer vision and deep learning technologies face challenges in efficiently determining aesthetically pleasing cover images and accurately scoring images based on various criteria.
Innovation Solution
The proposed solution involves a method and apparatus for obtaining a cover image by cropping an original image into multiple images, assessing each cropped image's aesthetic score, and selecting the target cover image based on these scores. Additionally, an image scoring model is trained using reference and sample cropped images, with parameters such as coincidence, slicing detection, and aesthetic scores being considered to determine the target score for each image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If multiple cropped images are generated and evaluated using aesthetic scores, then the quality of the selected cover image is improved, but the computational time and processing complexity increase
Solution Approach 1:
The patent performs preliminary cropping of the original image into multiple candidate cropped images before the selection process. By pre-generating these cropped images with different compositions and focusing areas, the system establishes a pool of candidates that can be quickly evaluated using the image scoring model, rather than generating and evaluating them in real-time during selection
Solution Approach 2:
The patent replaces manual or rule-based cover image selection with an automated image scoring model based on deep learning. The model automatically evaluates aesthetic scores, composition quality, and subject prominence of multiple cropped images, substituting mechanical/AI processing for human judgment and significantly improving both selection quality and efficiency
2Measurement precision
If the image scoring model incorporates multiple evaluation parameters (coincidence parameter, slicing detection, aesthetic score), then the accuracy of image scoring is improved, but the model complexity and training difficulty increase
Solution Approach 1:
The patent segments the image evaluation process into three independent modules: coincidence parameter calculation (measuring alignment with reference images), slicing detection (detecting cut-off objects), and aesthetic score evaluation. Each module processes specific aspects independently and outputs are aggregated to form the final comprehensive score, making the complex evaluation manageable and interpretable
Solution Approach 2:
The patent designs a universal image scoring model that simultaneously performs multiple evaluation functions: aesthetic assessment, composition quality evaluation, and subject prominence detection. This multi-functional model replaces the need for separate specialized models for each evaluation aspect, reducing overall system complexity while maintaining comprehensive evaluation capability
Data Source
AI summary
A method for obtaining a cover image includes: obtaining a plurality of first cropped images of an original image corresponding to a candidate resource; obtaining an aesthetic score of each of the plurality of first cropped images; and determining a target cover image of the candidate resource from the plurality of first cropped images based on the aesthetic score of each first cropped image.