Real-time preprocessing of large field of view images using optical spatial filtering

Abstract

Cueing or locating objects of interest in a large field of view (FOV) image is a preprocessing requirement for recognition systems. Cueing operations generally utilize computationally intensive digital methods that can limit system performance, especially for large images. This digital bottleneck can be overcome by taking advantage of the speed and massive parallelism of optical cueing techniques that utilize spatial filtering to locate objects of interest. A well established design for optical filtering is the coherent imaging architecture with an accessible filter plane. This architecture (used for Vander Lugt correlation) is ideal for any type of spatial frequency filtering, and it can be implemented by using liquid crystal televisions as gray level filters and input image transducers.1, 2 This paper examines real-time optical gray level spatial filtering techniques for object location. Bandpass gray level windows can be designed by using Hamming or other apodizing methods to reduce the image ringing caused by binary bandpass filters. Filter optimization results in fewer object location false alarms and more accurate cueing. Gray level filters based on object scale have been used successfully to locate objects of interest in large FOV gray scale images.