Image Target Recognition Method

Target recognition is very important in industry and military affairs. Now there are two basic target recognition strategies, such as processing from bottom to top, which is called data-driving method, and processing from top to bottom, which is called knowledge-driving method. The former begins with low layer processing for example, general segmentation, label and feature extraction, then judges whether the feature vector extracted from the labeled area is in accordance with the feature vector of the object model. The latter firstly brings forward a hypothesis on probably existed feature, secondly proceeds with purposeful segmentation, label and feature extraction, lastly judges whether the feature vector extracted from the labeled area is in accordance with the feature vector of the object model.This paper adopts the target recognition latter strategy to identify two typical man-made objects-bridge over river and rural roadway net. Based on others' researches, through having implemented a great deal of experiments, this paper has put forward own algorithms. All of them are robust, real-time and effective.This paper respectively introduces two existed methods on identifying bridge and roadway, and points out their advantages and disadvantages.During segmentation of the bridge image, this paper synthetically thinks over the advantages of self-adapted selection threshold and iterative selection threshold, and presents own self-adapted iterative selection threshold. What's more the variance of the infrared image is neatly applied to the visible image. At the time of beeline extraction, this paper mends Hough Transformation, and use novel method, which reasons the information gamed from the transformation field into the image field, to eliminate the inability to locate beeline of Hough Transformation.During treating with edges, this paper constructs a means, named curve pattern, to solve the problem of curve matching, and brings forward a thought of combining areas with edges to enhance identification reliability.