Research On Image Target Recognition Method Based On Shape

As an important part of the pattern recognition, image target recognition has great application value. Through tens years’development, it has wide application in military affairs, space search, medicine and video surveillance areas. Based on the demand of image targets recognition in remote sensing military reconnaissance, theories and methods of image targets recognition using shape information is researched.On the basis of the characteristics that single sementation methods can not get good sementation results in complex scenes images, two image segmentation methods are proposed, one uses region growing and active contour model methods and the other uses region growing and threshold methods; an illusive target filter method based on transcendental information is proposed. In two segmentation methods, first, glancing regions are gotten by region growing segmentation, and then threshold method or active contour model method is applied on targets’local image. Because of the use of targets’s local image data and the eliminatioin of the other image scenes data, the segmentation gets good sementation performance. Targets’transcendental information and target regions’characters are used in illusive target filter method. This method can filter most illusive targets.After investigation of existent image target characters distilling methods based on the shape information, many new character distilling methods are proposed, such as the ratio of target area to its protruded shape area, cirque division, curvature mesure, the distance of the center to contour pixel points. Those algorithms’implementation methods are given. The experiment results show that those characters have the property of RST invariability.Model matching recognition method using invariable characters is researched. An image target recognition method named multiple invariable character values change bound matching is proposed. Based on the property that the contour of segmented image targets has change, a slack contour region model matching image target recognition method that is based on shape arrangement matching is proposed. The experiment results show that the proposed algorithms get good recognition performance.