Research On The Screen Image Quality Evaluation Method Based On The Edge Feature Of The Gradient Domain

The content of digital images is no longer confined to natural scene images(NSIs)in intelligence information processing.In recent years,a kind of image generated by computer rendering on the screen of various terminal devices has gradually increased,which is called Screen Content Image(SCI).Research shows that screen images can produce different image features compared with natural images.In the process of image processing,it is impossible to avoid the degradation of image quality caused by the introduction of various distortion types,such as gaussian noise and compression.Most of the existing image quality assessment models are developed for natural images,but they can not accurately evaluate the perceived quality of screen images.Therefore,how to accurately evaluate the quality of SCI is very urgent.Based on the perceptual characteristics of the human visual system and the basic characteristics of the screen image itself,this thesis proposes an image quality assessment model(EFGD)based on the edge features in the gradient domain,which is targeted to the screen image luminance channel and chroma channel in the gradient domain.Three basic edge feature maps of edge sharpness map(ESM),edge luminance contrast map(EBCM)and edge chromaticity map(ECM)are extracted,and a gradient contour model which can better represent the edge space layout of the screen image is used.GPS)to measure the edge sharpness of the screen image.A new calculation method is used to capture the brightness and contrast variations(ie,edge brightness/contrast)of the screen image gradient map.The edge chromatice is represented by the color moment of the chroma channel in the gradient domain.Finally,the experimental results on two commonly used screen image databases verify the superiority of the model(EFGD),indicating that the model is closer to the subjective evaluation results than most existing image quality assessment models.The innovation of this thesis is:(1)Firstly be extracted edge features in gradient domain,which can well reflect the prominent visual changes in distorted SCIs.(2)Gradient profile model is implemented to measure the edge sharpness feature,which can better represent the details of edge in SCIs.(3)The proposed feature extraction of edge brightness/contrast also plays important role in improving the performance of the model(EFGD).(4)An adaptive weighting algorithm is designed according to the statistical distribution of images,which adjusts the fusion weights of the three similarity maps.