Optical multiple pattern recognition with a correlator using a single binary phase-only filter

Optical correlator-neural network hybrid system for many patterns recognition

“…It has also been shown that a filter can be designed so that a single correlator can perform many correlations [5] to identify a range of patterns over an input set. Furthermore it has also shown that an optical correlator can be trained so that a single correlator filter can detect as well as discriminate the input pattern sets by having coded outputs [1][2], known as multiplex filters. However, similarities in the input patterns and the choice codes in the output to represent those patterns can affect the performance of the filter.…”

“…Manivannan at el showed both theoretically and experimentally [1][2] that an optical correlator in a standard 4f system can be trained so that many patterns can be recognized with coded outputs using only a single binary phase-only filter ( referred to as a 'multiplexed filter'). In contrast, a conventional optical 4f correlator [4] which uses N matched filters sequentially for N-patterns recognition.…”

“…The hybrid system is trained by combining two algorithms, for the very first time for the authors knowledge, one for training the correlator filter (Direct Binary Search algorithm [1][2]) and one for the neural network layer (error back propagation algorithm [1,3]). The combined algorithm will be presented in detail in another paper.…”

“…The processing time of the optical correlator is just the time taken for light to travel the length of the 4f system (typically few nano seconds) and is independent of the size of the information it process, making it possible for real-time application like in [6]. The input pattern is encoded with random binary overlay mask to increase the capacity of the correlator filter as shown in [1][2] by making effective use of the spatial frequency of the multiplexed filter. In this paper the hybrid system operating is demonstrated to recognise a set of sixteen English alphabetical characters.…”

Automatic detection of active sweat pores of fingerprint using highpass and correlation filtering

“…Correlation: Correlation is a well established theory to compare two signals [8]. Mathematical representation of the two-dimensional correlation of two 2D signals g(x,y) and h(x,y) can be given by the following equation:…”