Theory And Application Research On Rough Sets Facing To Object Detection

Object detection refers to ascertain the interested target existing or not in image and mapthe locations of it．Veracity and real time property is one of important evaluation indexes ofsystem．In general，object detection includes two parts：first，decide whether object exists inimage，then，if exit，determine the feature and location of object．It is meaningful to obtainthe desired object that is image segmentation for image matching and recognition．Now thewidely used algorithms were summarized such as thresholding segmentation，edge detection，region-based segmentation， clustering segmentation and so on． Thresholding imagesegmentation, due to its simplicity, easy implementation, and advantages of finding closeedges of objects, has been a classical and efficient technique in various applicationscommunities．It is difficult to select a proper threshold for image segmentation because that valuegradation make various regions have fuzzy boundaries，nearby gray levels roughly resembleeach other and values at nearby pixels have close relationship．In recent years，rough settheory and some developed theory have been introduced to the image processing，improvedsegmentation performance．On the basis of the theory of object extraction and rough set，image model andthresholding segmentation based on rough entropy are studied deeply．Combining with fuzzyrough set，S-rough set，several image segmentation algorithms are proposed for differentrequirements．Furthermore，a kind of edge detection algorithm is put forward based on imagemodels．The main contribution in this paper can be summarized as follows．(1) model for image representation: Interprets the whole gray level image to be a fuzzyset and defines two rough fuzzy sets approximating object and background of an image，respectively．The concept of S-rough set is introduced to represent images as S-rough sets bydefining migration rule．In order to adjust the transfer of singular points，images are furtherdescribed by variable precision S-rough sets in combination with the notion of inclusiondegree．(2) algorithm for thresholding segmentation: A kind of rough entropy that makes acompromise between object roughness and background roughness is established to determine the threshold of image segmentation．According to maximum entropy principle, the requiredthreshold for good performance of image segmentation is selected to be the threshold whichrough entropy is the maximum．Three thresholding segmentation algorithms based on imagemodels mention above is presented．(3) algorithm for edge detection: Viewing gray edge as the treated object，a method fordetermine the optimal threshold is proposed to obtain binary edge by using thresholdingsegmentation algorithms based on image model．By introducing “edge”characteristics inS-rough set image model，a algorithm of two direction S-rough set is established for edgedetection．(4) Experimental results show that the proposed algorithms are more effective andflexible，also restrain noise better.．...