Study On Microscopic Image Quality Assessment Based On Visual Perception

With the intensive research on the micro-world,various kinds of microscopy equipment are emerging.Scanning electron microscope(SEM)attracts the attention of the majority of scholars due to its high depth of field and ultra-high magnification.At present,SEM has been widely used in mining,materials,biology,and medicine etc.Due to the convenience of imaging,a routine microscopic experiment may generate a large number of pictures.However,in the imaging process,because of the judgment difference of different operators,different kinds of distortion,such as blur and noise which are the most common,will affect qualities of microscopic images.These distorted images can affect the researchers' analysis and judgment of samples.Therefore,high-quality microscopic images are important prerequisites for ensuring the accuracy of results.The image quality assessment(IQA)can select good quality images from amounts of images and help researchers save a lot of time for subsequent analysis.This thesis analyzes perceptual features of the human visual system based on Gestalt's principle of visual psychology.According to characteristics of visual perception,this thesis decomposes a micrograph into two parts which are cartoon and texture.Based on cartoon and texture decomposition,the thesis studies the objective assessment algorithm of microscopic images quality.For blurriness and noise in microscopic images,two no-referenced quality assessment algorithms are proposed.The specific research contents of this thesis are divided into the following three points:First,according to characteristics of visual perception,micrographs are decomposed into cartoon and textural parts.According to the principle of simplification and continuity in Gestalt's visual psychology,the visual perception system is more sensitive to changes in flat areas and edges of images.Based on the difference between the reductions of local total variation in different regions,the directional filter bank is used to decompose micrographs into cartoon and textural parts.Second,based on cartoon parts of micrographs,a no-reference micrographs quality assessment method for blurriness is proposed.Firstly,the cartoon part is mapped into frequency and spatial domains.And then,two methods are used to create spatial maps,in which one is based on the sum of local variation and the other is based on the maximum local variation(MLV);then the edge detection is performed on spatial maps based on MLV.The edge sparsity is used as the measure of blurriness.At the same time,the feature fusion of frequency and spatial maps based on the local variation is performed to obtain another measure of blurriness.Finally,two measures are normalized and final scores are weighted summation of normalized measures.Experiments show that the algorithm can accurately assess extents of micrographs blurriness,and objective scores maintain a high degree of consistency with subjective evaluation scores.Third,based on the decomposition of the cartoon and textural parts,a no-reference quality assessment method for micrographs noise is proposed.Traditional noise evaluation methods ignore the effect on IQA by visual perception system due to noise-induced image changes.In this thesis,pixels in the flat area are firstly located by cartoon parts of micrographs,and then the noise in the corresponding area is found in the textural area,thus avoiding the influence of the high-frequency information of micrographs on evaluation of noise.Finally,the variance of the obtained noise is used as the final score..Experimental results show that the algorithm can accurately measure the noise intensity and objective scores maintain a high degree of consistency with subjective evaluation scores.