Study Of Model-based Target Recognition

Image process and recognition technology are an interdisciplinary front Hi-Tech. In the past, 20 years, with the development of the computer and information science, it is deeper, more extensive that image process and recognition technology are developed. The methods of the target recognition can be divided into three big classes: target recognition based on knowledge, target recognition based on feature, target recognition based on model.Based on the conclusion of existing research fruit, this article focuses on the study of model-based target recognition. Because recognition based on knowledge is deficient in such aspects as experience, knowledge and cognitive model, we sets up degradative object model and makes a precise description to the object, thus the better one is discerned. In this paper, the following aspects has been researched:1. Some principled problems about the practical application of the methods to target recognition are discussed. The information processing peculiarity and method of target recognition based on degradative model is analyzed to solve the problems existing in the traditional target recognition.2. After analyzing various kinds of methods of edge detection, we select a method of edge detection.3. Hough Transform is a common method used to extract lines in binary image, but it cannot meet the requirement of real-time processing because of its high complexity in computation. In this paper, based on the basic principles of Hough Transform, a new algorithm is suggested to detect with high speed the vertical and horizontal lines in binary image.4. As to input characteristic figure very much heavy in testing originally, the neural network can't reach very good precision of discernment. The Snow learning architecture not only significantly improve learning efficiency and increase the dependability of the algorithm, but also the highest accuracy of target recognition.At present the test proves that the method of target recognition based on degradative model is feasible. And it can be expected to occupy the remarkable position in the research field of three-dimensional target recognition.