Correlating a template with video using an optical correlator

A common problem arising in optical processing is that problem dimensions often do not match the dimensions for the optical devices or optical architecture. Such is usually the case for time t in a video camera that provides continuous samples of x, y and time t. We wish to correlate incoming video data with a template video in real-time so we can see variations in time between the data and template. In block correlation we split the time sequence into segments that can be handled with the optical correlator. However, convolution of a data segment with a template leads to increased length relative to the input data. Because convolution in an optical correlator is computed in the frequency domain and sampled data is periodic in the frequency domain, the additional values are aliased and distort the result. There are two methods to remove errors due to aliasing: (a) the overlap and add (b) the overlap and save. We show how the optical correlator may be modified for the overlap and save method which seems to be simpler to implement than the overlap and add.