Spatially-integrating Spatial Light Modulators For Image Feature Extraction

Abstract

Optically-addressed Spatial Light Modulators (SLMs) have been THE critical active element in optical image processing systems since early days. OriginalIy designed to perform simple image transduction (incoherentto coherent conversion, wavelength conversion, amplification), the operations performed by the SLMs have grown steadily over the past several years. The culmination of this trend has been in the incorporation of sophisticated electronic processing with each pixel in the "Smart Pixel" devices. A price paid in increasing the operational capability is increased fabrication complexity and (for a fmed &vice area) reduced space-bandwidth product of the images to be processed. In this paper we present a modification of the basic optically-addressed SLM design enabling it to perform image feature extraction operations. We first describe the design modZication, then give device fabrication details, present experimental results and summarize. An optically-addressed SLM consists on a sandwich of planar layers of transparent conducting electrode, photosensitive materid, a dielectric mirror to isolate input and output optical signals, a liquid crystal material and another transparent conducting electrode with a bias voltage applied across the two outer electrode layers. Such a device can perform linear or nonlinear transfer operations point-by-point on the input image incident on the photosensor layer. For image feature extraction, it is important to perform neighborhood operations where one point in the output image depends on a neighborhood of the corresponding point in the input. This neighborhood operation can be achieved by introducing an additional pattemed electrode layer between the dielectric mirror and the liquid crystal film. The schematic diagram of the modified SLM is shown in Figure 1. Since the patterned electrode layer establishes islands of equipotential regions which in turn modulate the liquid crystal film, the output image is divided into corresponding regions of equal intensity. The output intensity of a given region is proportional to the input light intensity spatially integrated over that region, hence the name Spatially Integrating Spatial Light Modulators (SI-SLM). It should be noted that the region of integration is totally determined by the floating electrode geometry and hence is under control of the designer. In this paper we describe a SI-SLM in which the floating electrode is divided into concentric rings in one half plane and circularly oriented wedges in the other half. This wedge-ring geometry in photodetectors has been proposed and demonstrated for extracting position, scale and rotation invariant features from the power spectrum of an image [l]. Implementing a similar structure in an SLM allows these features to be input to a subsequent optical neural net image classifier. This advantage is even more significant when one considers an optical system that calculates these features locally via a lenslet array and a corresponding 2-D array of wedge-ring SI-SLM. Implementing similar structure in a detector arrays will lead to a severe multiplexing problem. The feasibility of the SI-SLM design concept was demonstrated with an amorphous Si ferroelectric liquid crystal binary SLM fabricated at University of Colorado, Boulder [2]. The hydrogenated a-Si film is deposited by plasma-enhanced chemical vapor deposition. The films are deposited in a p-i-n configuration to a total thickness of about 2 microns. The floating electrode layer consists of 500 angstrom thick chromium layer that is evaporated and pattemed using photoresist lift-off. This layer also serves as a reflecting layer. The thickness of the liquid crystal cell is 3 microns and the cell is filled with BDH SCE13 ferroelectric liquid crystal material. For proof-of-concept demonstration the device contained three wedges, three rings and a circular spot in the center calculating the DC component of the image power spectrum. The DC spot is 50 microns in diameter and the overall size of the device is 5 mm across. When this device is used in conjunction with a 300 mm lens, the spatial frequency components are divided into three bins for three rings (0 to 3 l/mm, 5.7-8.2 l/mm and 10.7-13 l/mm).