Optical supervised filtering technique based on Hopfield neural network

Abstract Hopfield neural network is commonly preferred for optimization problems. In image segmentation, conventional Hopfield neural networks (HNN) are formulated as a cost-function-minimization problem to perform gray level thresholding on the image histogram or the pixels’ gray levels arranged in a one-dimensional array [R. Sammouda, N. Niki, H. Nishitani, Pattern Rec. 30 (1997) 921–927; K.S. Cheng, J.S. Lin, C.W. Mao, IEEE Trans. Med. Imag. 15 (1996) 560–567; C. Chang, P. Chung, Image and Vision comp. 19 (2001) 669–678]. In this paper, a new high speed supervised filtering technique is proposed for image feature extraction and enhancement problems by modifying the conventional HNN. The essential improvement in this technique is to use 2D convolution operation instead of weight-matrix multiplication. Thereby, neural network based a new filtering technique has been obtained that is required just 3 × 3 sized filter mask matrix instead of large size weight coefficient matrix. Optical implementation of the proposed filtering technique is executed easily using the joint transform correlator. The requirement of non-negative data for optical implementation is provided by bias technique to convert the bipolar data to non-negative data. Simulation results of the proposed optical supervised filtering technique are reported for various feature extraction problems such as edge detection, corner detection, horizontal and vertical line extraction, and fingerprint enhancement.

[1]  M S Alam,et al.  Heteroassociative multiple-target tracking by fringe-adjusted joint transform correlation. , 2004, Applied optics.

[2]  Mohammad S. Alam,et al.  Feature extraction technique based on Hopfield neural network and joint transform correlation , 2004, SPIE Optics + Photonics.

[3]  K. M. Curtis,et al.  Optoelectronic implementation of a multifunction cellular neural network , 1996 .

[4]  Mohammad S. Alam.  Deblurring using fringe-adjusted joint transform correlation , 1998 .

[5]  Jun Li,et al.  EXPERIMENTAL INVESTIGATION OF A REAL-TIME NONLINEAR JOINT TRANSFORM CORRELATOR , 1994 .

[6]  X Yang,et al.  Optical wavelet transform by the phase-only joint-transform correlator. , 1996, Applied optics.

[7]  P Birch,et al.  Experimental implementation of a Wiener filter in a hybrid digital--optical correlator. , 2001, Optics letters.

[8]  Hidemitsu Ogawa,et al.  Optoelectronic architecture for cellular neural networks based on a new bias method and spatial coding scheme , 1997 .

[9]  Noboru Niki,et al.  Hopfield neural network for the multichannel segmentation of magnetic resonance cerebral images , 1997, Pattern Recognit..

[10]  Chuan-Yu Chang,et al.  Medical image segmentation using a contextual-constraint-based Hopfield neural cube , 2001, Image Vis. Comput..

[11]  B E Saleh,et al.  Optical implementation of the Hopfield algorithm using correlations. , 1990, Applied optics.

[12]  Jean-Louis de Bougrenet de la Tocnaye,et al.  Real-time demonstration of an on-board nonlinear joint transform correlator system , 1997 .

[13]  M A Karim,et al.  Optical fuzzy systems based on efficient implementation of a fuzzy associative memory. , 1998, Optics letters.

[14]  M A Karim,et al.  Optical implementation of cellular neural networks based on bias and joint correlation. , 1999, Applied optics.

[15]  R Gil,et al.  Non-linear filtering of ultrasonic signals using neural networks. , 2004, Ultrasonics.

[16]  D Psaltis,et al.  Optical implementation of the Hopfield model. , 1985, Applied optics.

[17]  B Javidi,et al.  Optical implementation of neural networks for face recognition by the use of nonlinear joint transform correlators. , 1995, Applied optics.

[18]  Mohammad A. Karim,et al.  Cellular neural network implementation using a phase-only joint transform correlator , 1999 .

[19]  Jzau-Sheng Lin,et al.  The application of competitive Hopfield neural network to medical image segmentation , 1996, IEEE Trans. Medical Imaging.