Features Prediction Of Typical Ground Targets And Research On Image Matching Algorithms

Infrared (IR) image matching is one of the key techniques of automatic target recognition (ATR). The IR target recognition performance depends on the corresponding research work on IR image matching algorithm and the technique of the reference map preparation.Currently, the reference maps are mostly visible image and the real-time images are IR images. There is a sharp intensity difference between visible and IR image, as well as different time IR images, because of the imaging theorems. There exists a problem of the reference map preparation, but there is no simple model or algorithm to change visible images to IR images. Exact intensity preparation or image synthesis is not necessary, because image matching algorithm may be based on feature images. Thus we can predict the feature image of the targets based on IR imaging theorems and characteristics. Aiming at the common problems, the paper does some studies of the techniques.In the first chapter, we first introduce the IR imaging theorems and characteristics, analysis all kind of main factors affecting the IR imaging. And we research on the changing low of the sun-radiation and the air-temperature, establish models of them.In the second chapter, we study the IR imaging model of targets, establish the thermal balance equation, then the surface temperatures of different targets background in different seasons or different time are acquired. We analysis the theorems of the Infrared sensor, then the infrared radiation images are acquired. By using the theorems of edge detecting, we predict the feature images of the targets background, which are the reference maps.In the third chapter, we introduce same image matching algorithms based on feature images, including Correlation matching algorithm, Hausdorff distance image matching algorithm, weighted Hausdorff distance image matching algorithm, shape matching by shape context. A method for image matching base on structure information is proposed. The experimental results show that this method has better performance with the advantage of low computational quantity.