Image quality assessment methods, systems, electronic devices, and readable storage media
CN114972305BActive Publication Date: 2026-05-26SHANGHAI WINGTECH INFORMATION TECH CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI WINGTECH INFORMATION TECH CO LTD
- Filing Date
- 2022-06-17
- Publication Date
- 2026-05-26
AI Technical Summary
Technical Problem
Existing convolutional neural network image quality assessment methods require a large number of subjective human evaluation samples, and their accuracy decreases when the amount of data is small.
Method used
By calculating the parameter differences of image training samples, a scatter plot is drawn and divided into multiple regions. Traditional algorithms are combined with convolutional neural networks, and training is performed in these regions separately, reducing the workload of subjective human evaluation and improving accuracy.
Benefits of technology
While reducing the number of training samples, it improves the accuracy of image quality assessment, especially when the amount of data is small, thus improving the training effect of convolutional neural networks.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN114972305B_ABST
Abstract
This application provides an image quality assessment method, system, electronic device, and readable storage medium. The assessment method includes: calculating the parameter difference of image training samples; performing subjective human evaluation on the image training samples to obtain human scores; plotting a scatter plot based on the parameter difference and the human scores of the image training samples; dividing the image training samples into multiple different regions based on the scatter plot; and inputting the data of each region into a convolutional neural network for training and fitting, thereby obtaining multiple trained quality assessment models. This application utilizes a combination of traditional algorithms and convolutional neural networks. The data to be trained is divided into different regions using the parameter difference calculated by the traditional algorithm and then fed into multiple convolutional neural networks for training. Since the parameter difference calculated by the traditional algorithm already closely matches subjective human evaluation, the accuracy of the trained neural network is relatively high.
Need to check novelty before this filing date? Find Prior Art