Researches Of Image And Signal Processing Methods In Vision Rehabilitation

Vision is the most important way for human beings to obtain the information ofthe world.Vision abnormality affects the quality of life of sufferers.Therehabilitation of vision is the requirement of improving their living standard.However,it faces many difficulties and challenges.Blindness is the most severe kind of vision abnormality,while color visiondeficiency is another kind of vision abnormality that has a wide influence range.Therehabilitation schemes for these two kinds of vision abnormality are still in theexploratory stage,and hence have many problems to be solved.This dissertation focuses on image and signal processing methods in visionrehabilitation schemes for color vision deficiency and blindness.Studies have beencarried out in following aspects.(1) To simulate the color perception of dichromats and avoid the highcomputation cost of conventional methods,a new simulation method is proposedbased on the matrix transformation.The method is easier and quicker than existingmethods and thereby fits for the processing of large amount of images.To make upthe absence of perception simulation for anomalous trichromats,a weighted averagemethod is proposed to simulate the color perception of anomalous trichromats withdifferent types and severities.(2) During the investigation of color vision rectification,an artificial neuralnetwork model is proposed to rebuild response signals of abnormal cones by training,which indicates that normal cone response signals can be rebuilt from abnormal coneresponse signals.To solve the problems of low color utilization rate and adaptabilityin previous color transformation methods for color vision rectification,aself-organizing color transformation method is proposed.The method adaptivelychanges colors of a scene into discernable ones for dichromats,so as to enhance theircolor discrimination.(3) Studies for the edge detection have been made according to the requirementof vision rehabilitation project for the blind.To alleviate the performancedegradation of the edge detection under noisy environments,two new features,named as color distribution similarity and orientation consistency respectively,areexploited in the edge detection.Accordingly,a feature fusion method is proposed for the edge detection of color images.The method has a better performance thanconventional ones under noisy environments.When calculating features,general histogram similarity measures havedisadvantages when several kinds of deformation happen to histograms.A novelsimilarity measure is proposed based on the average translation of histogram bins.The measure of average translation of bins outperforms general measures in thecondition of histogram deformation caused by illumination variance or noise of theimage.(4) Studies for the contour extraction and separation have been made accordingto the requirement of vision rehabilitation project for the blind.A novel method isproposed to separate individual object contours in an image by clustering edges withdifferent local features.The method adopts the k-means clustering and the SOMrespectively to group edge pixels with similar features,in order to extract contours,separate different object contours,and separate contours from the background.Based on this study,a self-organizing feature map model is proposed tosimulate the visual processing mechanism for color scenes.The model cansimultaneously fulfill multiple tasks such as the figure-ground separation,the edgedetection,the contour extraction,and the object separation.It helps for furtherinvestigating the signal processing mechanism of the visual system.(5) To find appropriate neural stimulating methods in the vision rehabilitationproject for the blind,the neural encoding mode for the visual information has beeninvestigated.A coherent oscillation model for the contour extraction and separationis proposed.This model can extract and separate contours of different objects bysynchronization and asynchronization of neural oscillation.It helps to promote theinvestigation on the neural encoding mode for the visual information,and alsoprovides suggestions for the selection of neural stimulating methods in the visionrehabilitation project.All proposed methods in this dissertation are based on characteristics andmechanisms of vision system.They have both good versatility for different imageprocessing applications and specific advantages for vision rehabilitation schemes.Itis expected to provide help to the vision rehabilitation for color vision deficiency andblindness,and thus bring new hope to sufferers.