Optics & Laser Technology. Vol. 28, No. 2, pp. 13-82, 1996 Copyright Q 1996 Elsevier Science Ltd Prmted in Great Britam All rights reserved 0030-3992/96 $15.00 + 0.00 0030-3992(95)00077-l ELSEVIER ADVANCED TECHNOLOGY Optical-digital processors for morphological and rank order filtering T. SZOPLIK, M. GEDZIOROWSKI Non-linear rank order and morphological filtering can be achieved in optical-digital processors. In these processors, all convolutions are performed in inherently parallel optical correlators. Arithmetic and logic operations are made digitally. Due to the threshold decomposition concept, grey scale images are sequentially treated slice by slice. The optical-digital method of local histogram calculation within both binary and weighted neighbourhoods allows local non-linear operations. We derive rank order and morphological filters from optically calculated convolutions. Several configurations of optical convolvers are discussed. Further improvements in the technology of spatial light modulators and encoded light sources are needed before these processors’ practical uses will appear. KEYWORDS: correlators, filters, image processing, optoelectronic processors Introduction Information processing networks of the next generation will connect local area networks of particular users with all possible sources of information such as, for example, national libraries, satellite image libraries, and many others. Links between particular points of the network will be realized by means of copper cable nets on the lowest short-distance level, fast optical fibre nets, high- speed optical fibre backbone nets, cellular wireless links, and communication satellite-to-ground station mutual links. The choice of the lowest level short-distance, either cable or wireless, connections will depend on the necessary data transmission rate. On different levels of the information network, signals will be transmitted in analogue or digital form. It is quite obvious that, in yet undesigned information systems of the future, there will be a demand for image and image-format data processors based on new photonic devices. We can expect that one of the necessary processors will be an optical-digital processor of image-format data and will play the role of a smart digital-to-analogue interface. The processor will treat once-received rough image-format data of considerable size according to user needs. It can have the form of a computer-controlled optical correlator that is built in a free space architecture and contains two electronically controlled input data arrays. The authors are at Warsaw University, Faculty of Physics, Institute of Geophysics, ul. Pasteura 7, 02-093 Warsaw, Poland. Received 1 December 1994. Revised 9 May 1995. Photonic devices for convolution kernel input can have the form of an encoded light source, such as an independently addressable vertical cavity surface emitting laser diode array. A photonic data input device can have the form of an electrically addressed spatial light modulator (SLM) made with ferroelectric or smectic liquid crystal over silicon technology. Depending on the configuration, optical-digital processors of this kind can work with both coherent and non-coherent illumination. Both types are well known and were studied with several specific applications in mind. During the last 50 years, the attention of the information optics community was attracted by Fourier optics’. This was a result of the feasibility of phase operations, which came to be used with holography, the availability of sources of coherent light, and such concepts as matched and phase-only filters. In a two- lens coherent 4fcorrelator, matched and pure phase filters are placed in the intermediate Fourier plane. As a result, correlation and convolution of an input image and a filter impulse response as well as filter operations on complex functions are possible. The two-dimensional Fourier transform represented in Cartesian coordinates is space variant; that is, in spite of the position of an input function, its optical Fourier spectrum is always centred on the 4f system axis. Due to the Fourier transform space variance, no mechanical movement of correlated functions is necessary. This fact stimulates the use of 4fcorrelators for pattern recognition purposes. 73