A Siamese Network-Based Contrastive Learning Image Quality Assessment Method

Abstract

The invention relates to a comparison learning image quality evaluation method based on a twin network. First, the image to be trained is normalized by local contrast, then divided into image blocks, and image pairs are generated; second, the structure of the twin convolutional neural network is designed, and the image quality assessment model is trained using the designed network; finally, Divide the image to be tested into image blocks and generate image pairs. Use the trained model to predict the quality of all the image pairs to be predicted, obtain the quality ranking of all images, and obtain the quality score of each image according to the ranking. The method of the present invention proposes to convert the image quality evaluation problem into the quality comparison problem between image blocks, and use the pairwise comparison between image blocks to obtain the quality score of each image by counting the results of comparison between each image and other images, which can Significantly improves the performance of unreferenced image quality assessment.

Description

technical field [0001] The invention relates to the fields of image and video processing and computer vision, in particular to a comparison learning image quality evaluation method based on twin networks. Background technique [0002] Digital images are particularly important today when information technology is highly popularized, but images are often distorted in daily applications, such as in the process of image acquisition, compression and transmission. In order to apply digital images better, image quality evaluation becomes particularly important. With the development of convolutional neural networks, many researchers began to use convolutional neural networks for no-reference image quality evaluation. At present, many unreferenced image quality assessment algorithms based on convolutional neural networks have been proposed. For example, Kang et al. applied a shallow convolutional neural network to unreferenced image evaluation, and its performance was improved comp...

AI Technical Summary

Problems solved by technology

The previous unreferenced image quality assessment method based on convolutional neural network mainly solves this problem through two methods. The first method is to divide the image into blocks, and each image block uses the score of the complete image as the label. The quality of some parts is different, and it is not accurate to use the complete image to mark different blocks

The second method is to use the full-reference quality assessment method to annotate the image. The defect of this method is that the performance of the algorithm directly depends on the performance of the full-reference image quality assessment.

Images