Research On Image Enhancement And Quality Assessment Based On Biological Visual Perception Mechanism

The human visual system is an intelligent and efficient visual information processing system.Therefore,developing more effective image processing applications by simulating biological visual perception mechanism has attracted a lot of concern.The computational model established is focused on the content image enhancement processing,including image enhancement,denoising for enhancement image,and image enhancement quality assessment.Based on physiological-oriented and task-oriented approach,this dissertation provides new computational models and effective implementation methods for image processing applications such as image enhancement,image denoising,and image quality assessment by simulating the receptive field mechanism of biological vision and the perceptual characteristics of the visual system.In the first part of this dissertation,aiming at solving the problem of color distortion in current color image enhancement methods of the fundus,an improved color constancy Retinex enhancement method is proposed.This method firstly extracts the luminance channel,performs multi-scale Retinex processing on the luminance channel,then performs image mapping through an improved gain/offset algorithm and a color restoration method,and finally recovers the red channel,which has luminance information,according to the characteristics of the fundus color image.In order to vertify the validity of the method,the proposed method was compared with classical enhancement methods such as multi-scale Retinex(MSR),multi-scale Retinex with color recovery(MSRCR),histogram equalization(HE),Contrast Limited Adaptive Histogram Equalization(CLAHE)on the DIARETDB0 fundus image database by subjective and objective evaluations.The experimental result shows that the proposed method has better processing effects than the classical methods in terms of color protection,improvement of blood vessel contrast and enhancement of image details,which are of great significance to further identify the fundus image.In the second part of this dissertation,in order to solve the problem of increase of noise after nighttime color image enhancement,an image denoising method based on visual receptive field characteristics is proposed.Based on the results of visual neurophysiological studies,the proposed method simulates the response characteristics of the receptive field to fulfill the task of image denoising.It firstly transforms the color space,then detects the noise and depresses it by the adaptive ON/OFF receptive fields which are dynamically adjusted according to the size of the noise.The experimental result shows that the proposed method effectively suppresses noise,well preserves the textures and edges detail information,and greatly improves computational efficiency.The proposed method is superior to classical methods in terms of objective quantification indicators such as peak signal-to-noise ratio and mean-square error,solves the practical problem of the increase of noise amplification after nighttime color image enhancement,and further perfects the image enhancement method proposed in the previous section by expanding its application field.In the third part of this dissertation,in order to solve the problem of lacking of uniform and objective image assessment for the image enhancement methods,a novel image enhancement quality assessment method is proposed based on the human visual perception.Firstly,the characteristics of influencing the visual quality for an image are analyzed and described,then we present the visual feature metrics and the corresponding mathematical models.Finally,based on the integration of these visual feature metrics,we get the image enhancement quality assessment,called the VPMI(Visual Parameter Measurement Index).This method was verified using the popular assessment database Image Quality Assessment Database(LIVE),and the results indicated that it is highly competitive with respect to state-of-the-art methods,and the VPMI has a low computational complexity,which makes it promising to implement in real-time image assessment systems.Finally,we applied the method to assess the results of the image enhancement,which shows that the enhancement algorithm in the previous section is better than the other traditional algorithms.Furthermore,this part presents a method to detect the distortion of the enhancement image.The distortion here refers to the distortion degree of the inherent characteristics of the image.Based on the analysis of the inherent properties of images,we propose a new image distortion assessment framework consisting of three components(including the information distortion,constituent distortion and color distortion)to assess the image distortion rate for image enhancement.We compared the distortion rates of different image enhancement algorithms under the different conditions,which the experimental results demonstrate that any enhancement method can improve the image quality,whereas at the cost of image distortions.It is necessary to keep a moderate balance between image distortion and image quality when they assess the enhanced images.The results showed that the proposed distortion rate could provide an additional objective basis for image enhancement assessment method.Finally,we applied the method to assess the distortion rate for the image enhancement methods in the previous section,which shows that the enhancement algorithm in this research has lower distortion rate,especially for color distortion.