No-reference image quality assessment method based on convolutional autoencoder network

Abstract

The invention discloses a no-reference image quality evaluation method based on a convolutional autoencoder network. The specific steps of the invention are as follows: construct a convolutional autoencoder network; construct a fully connected neural network; generate a pre-training set, a training set and a test set; Train convolutional autoencoder networks and fully connected neural networks; perform quality evaluation on distorted images in the test dataset. The present invention uses the convolutional self-encoding network to encode the non-reference image and its image blocks respectively, uses the fully connected neural network to extract the global semantic feature and the local distortion feature from the encoding of the no-reference image and its image block respectively, and combines the two This kind of feature fusion uses the fully connected neural network to map the fused features into perceptual quality scores, which has the advantage that the evaluation results are more in line with human subjective feelings.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to a no-reference image quality evaluation method based on a convolutional self-encoding network in the technical field of digital image processing. The invention can be applied to objectively evaluate the perceptual quality of digital images without original reference images, so as to ensure the validity and accuracy of acquired digital image data. Background technique [0002] In the process of imaging, transmission and storage, digital images are affected by the optical system, compressed transmission and other factors, which will eventually cause the image obtained by the terminal to suffer from various image quality degradation problems such as compression distortion, Gaussian noise, and blur. The perceived quality of images is an important index to compare the performance of various digital image processing algorithms and the parameters of digital image imaging ...

AI Technical Summary

Problems solved by technology

The disadvantage of this method is that only the image block is used as the input of the network, and the integrity of the semantic content of the image is not considered, so that the accuracy of the evaluation result of the trained model is not high.

The disadvantage of this method is that the method uses manually extracted natural statistical characteristics NSS features to perform score fitting, which makes the extracted features less sensitive to image quality, resulting in evaluation results that cannot better conform to human subjective feeling

Images