Research On Image Quality Evaluation Method Based On Feature Extraction

In the information age,vision has become one of the main ways for people to obtain information.As an important part of visual information,images provide us with a wealth of information resources.High-quality images can not only satisfy human vision's pursuit but also provide more accurate information for research in many fields.However,various types of distortions are introduced in the process of image acquisition,storage,processing and dissemination.The quality of the image will be poor and the information will be inaccessible for us.Therefore,how to evaluate image quality quickly and accurately is very important,and it has also been widely noticed by many scholars in recent years.This thesis studies the quality assessment of Gaussian blurriness images and DIBR-synthesized images.The main research content includes the following two aspects:(1)Based on Ada Boost BPNN,a non-reference quality evaluation method is proposed for Gaussian blurriness images.We firstly extract four shallow features:ambiguity,roughness,linearity and clarity.Then,the VGG19 in deep learning is used to extract five deep features.The shallow features and depth features are combined through Ada Boost BPNN.Finally,the quality evaluation model is trained to evaluate the quality of Gaussian blurriness images.The experimental results show that the proposed method has more advantages than most of the existing methods,and has a high consistency with subjective evaluation results.(2)Based on hole and expanded regions,a quality evaluation method is proposed for the DIBR-synthesized images.We consider two representative distortion types,i.e.,hole regions and expanded regions.Firstly,the watershed algorithm is used to extract and label the hole regions.The dispersion and area of the region are used to design the hole features quality score.Then,the contour difference between the distorted image and the reference image is calculated to extract expanded features.The area of the region is used as the quality score of the distortion of the contour amplification.Finally,the two quality scores are pooled to gain the final quality evaluation score.The experimental results show that this method has high evaluation accuracy and it is consistent with the subjective perception of human eyes to the distorted image synthesized by DIBR.