ABSTRACT
In image recognition, it is desired that the receiver detect a known reference target in the presence of noise [1-7]. Also, the receiver should be able to detect the target even when the target is distorted. Possible sources of distortion are rotation, scaling, and changes in illumination. Numerous methods and algorithms have been developed to detect a distorted version of the target [8-23]. In one such method, a weighted superposition of distorted training true class targets to form a composite matched filter. However, many variations [8-16] of matched filters are proposed to allow distortion tolerance. There is also a vast amount of literature for pattern recognition using neural networks [18] and classification methods [17,18]. However, neural networks are more suitable for target classification rather than detection.
