Abstract

The rotating kernel min-max transformation (RKMT) has been proved useful for directional image processing and particularly well suited for optoelectronic implementation.^1,2 The method compares extremely well with the best competing algorithms we have found. A related processing operation can be used to obtain object signatures that are independent of the angular orientation of the 2-D object and that are largely independent of the object scale.² In this method, the input is convolved with a long, narrow kernel, and the maximum (MAX) and minimum (MIN) values of the convolution are stored. A typical output is obtained by some function of MAX and MIN. Basic RKMT operations are implemented by incorporating an optical convolver with a CCD camera as a detector on the output plane. The effects on performance of the kernel profile and relative dimension to directional features to be processed are discussed. The RKMT processing is also affected by the angular resolution. Thus the effect of the resolution is discussed for various kernel profiles, including rectangular and Gaussian profile. Experimental results will be discussed both qualitatively and quantitatively.

© 1992 Optical Society of America