Abstract

Matched spatial filters provide the highest output signal-to-noise ratio in pattern recognition applications but have not become practicable due to the complex nature of the spatial filters required. Lately much research effort has been devoted to methods of avoiding complex filters. This research has suggested binary and ternary real valued approximations to the phase-only filter. Dickey and Hansche¹ suggested a quad-phase-only filter that does not suffer from having the simple symmetry of real valued filters and can have improved signal-to-noise ratio. The authors suggest a complex ternary matched filter (CTMF). This CTMF filter function has the form: where I_R1 and I_R2 are support functions, s_r and S₁ are the real and imaginary parts of the object function Fourier transform, and β is the threshold line angle. The algorithm introduced by Kumar and Bahri² can be used to optimize this filter. The optimized CTMF produces signal-to-noise ratios very near the optimal phase-only filter and does not suffer from having the simple symmetry of real valued filters.

© 1989 Optical Society of America