Research On Image Quality Assessment Method

In daily life,people are receiving all kinds of information from the world all the time,80% of which are presented to us in the form of images.It is self-evident that image is of great importance to us.Similarly,the research in the field of image quality evaluation is equally important.Nowadays,in image quality evaluation methods,we often use objective evaluation methods to test distorted images.It has the advantages of small computation and easy access.This paper mainly introduces several typical full reference methods in objective evaluation methods,and analyses the performance of several full reference evaluation algorithms.Experiments show that,compared with other algorithms,feature similarity(FSIM)has better evaluation effect on different types of distorted images.On this basis,two Reference-No image quality assessment methods((SP_FS and WN_FS))based on FSIM are proposed.The core idea of the two methods is to denoise the distorted images of different distortion types and get the constructed reference images.Then the distorted images and reference images are processed by FSIM.Aiming at distorted image with salt and pepper noise,this paper adopts median filtering and FSIM-based no-reference image quality evaluation method.The image samples are three images provided by LIVE(release 2)image library.And each image is denoised with 10 groups of distortions.The denoised image is taken as the image to be measured,the median filtered image is taken as the reference image,Finally,the FSIM of the two images is calculated to measure the quality of the image to be measured;Aiming at the distorted image with Gauss white noise,this paper adopts the method of image quality evaluation based on FSIM and Gauss filtering.Three sets of white noise images provided by LIVE(release 2)image library are used as image samples,and they are used as reference images,The image filtered by Gauss is taken as reference image,Finally,the FSIM of the two images is calculated to measure the quality of the image to be measured.The experimental results show that compared with other full reference image evaluation methods,SP_FS and WN_FS evaluation method have some improvement in the performance indicators of the representation algorithm,Compared with the existing non-reference evaluation methods,the improved algorithm has the advantages of simple calculation and accuracy.The two evaluation methods have good evaluation effect on salt and pepper noise and Gauss white noise respectively,and have good consistency with the subjective perception of human eyes.