A deep learning image evaluation method for video transmission quality

Abstract

The invention discloses a deep learning image evaluation method oriented to video transmission quality. The method includes: constructing a Siamese neural network for evaluating image quality to extract features, and performing feature extraction on target image block feature maps and reference image block feature maps. Fusion, design image quality evaluation score calculation method and loss function to update network parameters; set twin neural network training hyperparameters, including learning rate, learning decay rate, maximum number of training steps, learning rate decay steps, batch training volume, training Image size, single image sampling amount, data set (training set, verification set, test set) distribution, etc.; design the image area segmentation and block selection scheme of the target image and reference image.

Description

technical field [0001] The invention relates to the field of image quality evaluation, in particular to an image quality evaluation method based on deep learning. Background technique [0002] Video images will have a great impact on the final transmission quality due to various reasons during the transmission process, so it is very important to correctly evaluate the image transmission quality. Many of the existing image quality evaluation techniques rely on the human visual system or computational models of natural image statistics, or even human evaluation. The correlation between the former evaluation score and the subjective average opinion score of the human eye is sometimes relatively low, and it is prone to inconsistency in judgment; while the latter is inefficient, and it is prone to fatigue errors when working for a long time, which affects the accuracy of judgment. Efficient, accurate and intelligent methods for evaluating video image transmission quality have im...

AI Technical Summary

Problems solved by technology

The correlation between the evaluation score of the former and the average subjective opinion score of the human eye is sometimes relatively low, which is prone to inconsistent judgments; while the latter is very inefficient, prone to fatigue errors after long-term work, and affects the accuracy of judgment. Efficient, accurate and intelligent methods for evaluating the quality of video image transmission have important practical significance

Images