No-Reference Image Quality Evaluation Method Based on Hierarchical Feature Fusion Network

Abstract

The invention discloses a reference-free image quality evaluation method based on a hierarchical feature fusion network, which mainly solves the problems of low precision and slow speed of the prior art. The implementation scheme is as follows: 1. The reference image is selected from MSCOCO data set and the polluted image database is established by adding noise; 2. Simultaneously de-averaging andprune that training set image and the t set image; 3. According to the hierarchical processing mechanism of human visual system from local features to global semantics, a hierarchical feature fusion network model is designed for end-to-end joint optimization. 4, train that hierarchical feature fusion network model by using a training set and a t set; 5, carry out de-averaging and cutting process on that images to be evaluated, and inputting the process images into a trained hierarchical feature fusion network model to obtain an image quality prediction score; The invention greatly improves theaccuracy and speed of no reference quality evaluation, and can be used for image screening, compression and video quality monitoring.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a reference-free image quality evaluation method, which can be used for image screening, compression, and video quality monitoring. technical background [0002] With the rise of the Internet and mobile terminals, massive image and video data are generated every day. Unfortunately, in the stages of image and video generation, processing, transmission and storage, various distortions will inevitably occur, so automatic evaluation of image quality becomes indispensable. Subjective quality evaluation came into being and has been widely used in the field of image and video. [0003] In the past ten years, a variety of image quality assessment algorithms with superior performance have emerged. According to whether the reference image is available, these algorithms are divided into three types: full reference image quality assessment algorithm, partial reference image quality...

AI Technical Summary

Problems solved by technology

However, this type of algorithm cannot be jointly optimized because feature extraction and quality prediction are separated, which greatly reduces network efficiency.

Although these networks have achieved great success, they still have disadvantages: 1) Although feature extraction and quality prediction can be jointly optimized for end-to-end networks, they only use the last layer for quality prediction, and

Images