Image Measurement Research For Automatic Target Recognition Performance Evaluation

As the input of Automatic Target Recognition ATR algorithms performance evaluation system, the image measurement is the very important part of ATR performance evaluation. Whether image measurement indexes accurately relates with algorithm performance will affect directly evaluations. According to the image measurement problem in ATR algorithm performance evaluations, this paper addresses how to characterize the properties of images effecting the ATR system or algorithm performance, with the basic principles for measuring images.In this paper, the global and local image measurement indexes based on one-second statistics of gray levels probability are introduced, and the usage conditions and shortcoming of the target background contrast TBC are analyzed as a stress.This paper describes the notion of the clutter, analyzes the affect of image clutters to ATR algorithms in detail and expounds the significance of image texture features to characterize images. In this paper, the texture analysis method based on gray level co-occurrence matrix is adopted, and this method is used to characterize the image clutters. Two image texture measurement indexes based on depending-target global clutter (ICOM) and target background metric (TIC) are proposed to measure images, the result of testing shows that s TIC has very good monotonic relation with the segmentation algorithm performance, efficiently gets over the shortcoming of TBC, and TIC is a fairly robust indicator of segmentation algorithm performance and better suited than TBC.At last, this paper gives a pilot study of the image database measurement. The meaning of image database measurements is discussed. Combined with the properties of image database, several image database measurements are proposed, and support a research direction for advance work.