Multi-scale image quality detection method based on convolutional neural network

Abstract

The invention discloses a multi-scale image quality detection method based on a convolutional neural network. In the training phase, it builds a convolutional neural network, which has eleven internal network layers, of which three layers of convolution are used for encoding, three layers of deconvolution are used for decoding, and the remaining seven layers are used to deepen the network, allowing the network to learn more High-level abstraction, the network also includes two skip layers and three multi-scale output layers; the original distorted image is preprocessed accordingly, input to the network for training, and the input data is mapped to the structural feature similarity map corresponding to the original image; The network parameters are updated iteratively through the directional propagation algorithm to obtain better network parameters; in the test phase, the features of the test image are extracted through the trained network, and the overall quality result of the test image is obtained through significant weighting. The invention does not use human subjective scores for supervision, and there is no training error caused by human subjective factors, so that the prediction result can be more objective and accurate.

Description

Multi-scale image quality detection method based on convolutional neural network technical field The present invention relates to a kind of multi-scale image quality detection method of deep learning, especially relate to a kind of based on convolution Image quality detection method with neural network. Background technique The detection of objective image quality is a fundamental problem in computational vision, and image capture devices vary in lighting conditions, exposure, light Circumstances, susceptibility to noise, and limitations of the lens can create image artifacts that can result in blurred images. using compression When the image is processed by the method, it will also cause the loss of image information and lead to the distortion of the image. In today's era of big data, no It is possible to detect image quality through artificial subjectivity, so an accurate image quality evaluation model is established to monitor Degradation of image quality, opt...

AI Technical Summary

Problems solved by technology

[0005] The existing no-reference image quality detection algorithm can obtain results similar to human subjective perception on various distortions, but the prediction accuracy will decrease when predicting all distortions, and the results for different databases are also different. The model The versatility is not strong

And there may be a variety of mixed distortions in images in real scenes, which increases the difficulty of image quality assessment

Images