Target Detection Technique Based On Probability Density Distribution And Fuzzy Entropy

Image target detection is not only a basic and essential stage in image engineering but also a key technology of machine vision. The extent of target detection is very broad, including static image segmentation and target extraction, as well as detection, recognition and tracking of moving targets, where good techniques for image segmentation are desired. Performance of an image segmentation technique has directly impact on efficiency of object detection in visual systems. Therefore, image segmentation is a core procedure of object detection. The study on it leads to important meaning to objective identification and target tracking.Since there exist inherent ambiguity of objects and their edges in image, image information can not be precisely characterized by classical mathematical theory. This is why traditional techniques of image segmentation are challenged. Due to the fact that fuzzy theory is a good implement of describing uncertainty and ambiguity, techniques of image segmentation based on fuzzy theory have been intensively studied in image analysis and processing. However, the difficulty to determine membership function of a fuzzy set leads to limilted techniques of image segmentation appeared. In addition, complex structure and large amount of computational complex of type 2 fuzzy sets make the utility of it constrained. Thus, the study of special case of type 2 fuzzy sets, namely interval-valued type-2 fuzzy sets (IVFS) has been extensively carried out. In this dissertation, a new IVFS fuzzy entropy measure is put forward and its properties are discussed on the basis of reviews of some common interval valued fuzzy entropies. Moreover, the proposed IVFS fuzzy entropy is applied to image segmentation. Experimental results show that it has good capabilities of image segmentation.The techniques of Image segmentation based on fuzzy theory mainly concerns fuzzy entropy and fuzzy clustering. However, there exist few achievements combining both of them in image segmentation. In this dissertation a Gaussian kernal function, as one kind probability density distribution function of image space, is introduced and successfully applied to micro-target and point object detection under low light and low contrast conditions. An image segmentation algorithm in combination with mathematical morphology and fuzzy C-means (FCM) based on probability density distribution is proposed. Its application to extraction of underwater objects is presented. Furthermore, a new technology of image segmentation combined fuzzy entropy with fuzzy clustering is developed. Simulated experiments show the validity of the proposed algorithm.The outline of this paper is depicted as follows.1. The theory of fuzzy entropy measure of interval type-2 fuzzy sets is studied. A new definition of entropy of interval type-2 fuzzy sets is proposed, its properties are investigated and its application to image segmentation is discussed.2. A method of edge detection of image based on probability density distribution is proposed. With this algorithm, micro-target and point object can be successfully detected.3. In combination with mathematical morphology, a new FCM algorithm of target detection based on probability density distribution is put forward.4. A new method to determine the upper and lower membership function of interval-valued fuzzy sets is developed and an algorithm of target detection combined the new entropy of interval-valued fuzzy sets with FCM is designed.