TABLE OF CONTENTS Optical pattern recognition is another equally important area of application of optical processing. Here optical correlation, spectrum analysis, or other "feature extraction" techniques may be used to determine key features of objects. Typically digital sequential computing devices would follow the feature extraction step to perform the object classification function, which is often based on using these key features for pattern recognition. This process is suggested by Fig. 12.1. However, parallel optical computing architectures have also recently been developed, which employ neural network concepts to integrate the feature extraction and object classification functions into one. These multilayer network architectures are amenable to optoelectronic solutions, where optical free-space coupling and nonlinear optoelectronic devices are married. Micro-optical arrays and 2-D matrix SLMs can be employed in building such architectures.
Specific examples of optical architectures will be discussed for calculating Mellin transforms, moments of a 2-D image, Walsh Hadamard coefficients, angular correlation features, and Radon transforms. Finally, the use of optical neural networks will be considered as a means to perform object classification.
The general problem of pattern recognition can be described in terms of a process summarized in Fig. 12.2. Although this is not the only way to look at pattern recognition, it is a common approach that delineates the subsystem functions and potential roles for optical processors. The sensor measures an object in at least two dimensions using various potential parameters, all of which are represented as complex or...
