Research On The Key Technologies Of Image Segmentation Based On Visual Perception

Image segmentation is an important research area with many applications such asbridge crack checking, intelligent vehicle navigation and infrared pedestrian detectionand so on, which has become the frontier and hot topic in the domain of computervision in recent years. Image segmentation technology includes image de-noising,image enhancement, image segmentation and other aspects, and has some difficultiesand challenges in image noise introduced by collection and transmission process, lowluminance, poor illumination etc, which make it a difficult and challenging task thatstudy robust, reliably segment, and has received much attention from researchers.Arming at these problems above, enlightening by human visual perception，centering on nature fuzzy of digital image. This paper deeply studied into imagede-noising based on multi-channel visual perception, image enhancement based oncontrast stretching and human color perception, fast image segmentation based on fuzzyRenyi-entropy, and ultimately forms a set method of image segmentation based onhuman visual perception and fuzzy theory.In the aspect of image de-noising, according to “over kill” the detail of waveletcoefficients and “under-kill” noise of existing methods, a new approach is proposed tofilter image noise based on local variance fuzzy wavelet. Considering distance andcontinuous contour of local region, define fan-shape local variance for rotation invariant.And on this basis, calculating wavelet coefficients membership by fuzzy function,realization fuzzy wavelet image de-noising. The proposed method effectively inhibitfuzzy edge contour, and acquire preferably effect.In the aspect of image enhancement, arming at Non-independent among colorimage channels, contrast stretched for gray image enhancement and human colorperception for color image enhancement are studied. According to the shortcoming ofgray image enhancement using traditional Beta function, extended Beta function, waspresented in this paper by introducing enhancement operator and new shape-controllingparameters, the selection mechanism of enhancement operator and shape-controllingparameters of extended Beta was put forward, the presented method retain more imagedetails, obtain satisfactory gray image enhancement result. On the other hand,considering Non-independent human color perception, according to the relationshipbetween human body perception and the corresponding real physical quantity depicted by Weber-fechner Law, the perception model of color is established, an improvinghistogram equalization algorithm that taking the visual perceptual characteristics ofcolor into consideration is proposed. On this basis, color image enhancement of humanvisual perception was put forward, the proposed method can effectively enhancecolorful image and make the detail information of image stand out.For the image segmentation, according to the nature fuzzy of digital image,defining new fuzzy Renyi entropy, following the maximum entropy principle, QuantumGenetic Algorithm QGA) is employed to accelerate the search of the optimalparameters of membership function, a new image threshold method based on fuzzyentropy is presented, this fast segmentation obtains good performance, and satisfies therequirement of real-time.At last, combined with research results above, an image segmentation approachbased on human visual perception and Fuzzy Theory is formed, the paper has carriedout the experiment using transportation scene as experimental data. The experimentalresults have certified that the algorithms can effectively eliminate the effect of noise,enhance the details of image, has strong robustness and great application values.