Abstract

The design of filters for optical pattern recognition is extensively studied for more than ten years [1] [2] [3] [4] [5]. Indeed, optical correlation implementation leads to specific constraints in comparisons with classical signal processing technics [6]. In particular, although the spatial matched filter is optimal for noise robustness, its limitations such as broad correlation peaks and low diffraction efficiency [3] are well known. Many different approaches have improved some of these characteristics and a lot of work is still devoted to find trade-offs between them [7] [8] [9] [10] [11] [12].

© 1991 Optical Society of America