Optoelectronic Training Architecture for Supervised Filtering Technique

Abstract

Supervised filtering technique is a kind of dynamic neural network that performs one filter mask and one bias value which are adjustable by a supervised learning algorithm for various types of applications. Training procedure for a supervised filtering technique needs much more computation and complexity for 2D data set compare to test procedure. Therefore, optoelectronic architecture has been developed for training procedure that presents high speed application possibility in hardware and incorporation availability with other optical systems. In optoelectronic architecture, we have employed Joint Transform Correlator (JTC) which produces same results with convolution operation when mask coefficients are symmetric. For image processing applications, mask coefficients should be optimized according to input and desired output images. This task is realized by supervised training stage based on the proposed optoelectronic architecture. In test stage, we have verified the proposed optoelectronic training architecture for image corner detection problem.