“…In this work a simple orientation operator based on horizontal and vertical Sobel filtering was employed, as described in [16]. Local orientation, represented by an angle between 00 and 3600, ja continuous quantity and can therefore be stored and processed as a floating-point number, as in [15]- [17]. Since, however, orientation filters in natural vision systems have a relatively large response width of the order [14], relatively course, discrete binary orientation channels can be used, which leads in turn to a significant simplification of any hardware implementation of the algorithm.…”