A system which can recognize compact targets from complex environments independently of objects’position

size and orientation is proposed. The system first selects threshold(s) automatically for the original image to get several binary images. In the simplest case

one threshold and

therefore

two binary images are got

one of which is got by setting 1 the pixels whose gray value is larger than the threshold

setting others pixel 0

and the other is completely opposite to this one. The binary images are used to determine the patterns’position .The system then uses the position data to extract rectangle areas from the original gray image. For every extracted area

a threshold is selected to get two binary images which are respectively transformed independently of position

size and orientation. Finally the geometrical moments are calculated for the transformed binary image to input to a BP as the image’s features to recognize it.