Interpretable Image Aesthetic Scoring via Scenario-Integrated Neural Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image aesthetic scoring methods cannot interpret the scoring results, limiting user understanding and experience.
Innovation Solution
An image aesthetic processing method that integrates scenario information into a neural network structure, using a convolutional structure set and an image classification neural network to generate an interpretable aesthetic scoring model, with a loss function calculation layer for improved robustness and accuracy, and applies this model in a dual execution environment for secure operation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If a depth model-based image aesthetic scoring method is used, then scoring can be performed, but the scoring result cannot be interpreted
Solution Approach 1:
The neural network is segmented into multiple functional modules: a backbone network for feature extraction, an image classification network for scenario identification, and a loss function calculation layer for scoring. This segmentation allows each module to perform its specific function independently, making the overall system both accurate and interpretable.
Solution Approach 2:
An image classification neural network is introduced as an intermediary component between the backbone network and the scoring output. This intermediary layer classifies image scenarios and provides interpretability by identifying what aspects of the image are being evaluated, while still enabling accurate scoring through the integrated architecture.
2Loss of information
If scenario information is integrated into the neural network, then interpretability is improved, but device complexity increases
Solution Approach 1:
The backbone network and image classification network are merged into a unified second neural network structure. This merging allows scenario information to be integrated seamlessly into the existing network architecture without requiring completely separate systems, thus improving interpretability while controlling complexity growth.
Solution Approach 2:
The second neural network is designed to perform multiple functions simultaneously: feature extraction from the backbone network, scenario classification through the image classification network, and aesthetic scoring through the loss function calculation layer. This multi-functionality reduces the need for separate dedicated components, managing system complexity.
3Reliability
If a dual execution environment is used for secure operation, then security is improved, but system complexity increases
Solution Approach 1:
The execution environment is segmented into a trusted execution environment and a non-trusted execution environment. The image aesthetic scoring model is deployed only in the trusted environment, which is isolated from external threats. This segmentation provides security by ensuring that the core scoring logic executes in a protected context while reducing overall system complexity through clear environmental separation.
Data Source
AI summary
This application provides an image aesthetic processing method and an electronic device. A method for generating an image aesthetic scoring model includes: constructing a first neural network based on a preset convolutional structure set; obtaining an image classification neural network, where the image classification neural network is used to classify image scenarios; obtaining a second neural network based on the first neural network and the image classification neural network, where the second neural network is a neural network containing scenario information; and determining an image aesthetic scoring model based on the second neural network, where output information of the image aesthetic scoring model includes image scenario classification information. In this method, scenario information is integrated into a backbone neural network, so that a resulting image aesthetic scoring model is interpretable. In addition, scoring accuracy of the image aesthetic scoring model can be improved by using the preset convolutional structure set.