Hardware-oriented cellular neural network software library investigation

Widespread use of visual data increases the importance of image processing in order to analyse images. It is obvious that the performance of image processing technics are directly related to accurate analysis. Thus, understanding behaviour of these technics plays an important role in solving potential problems. In this thesis, template library, which is a reference software library in image processing with cellular neural networks, is analyzed in terms of one of the hardware implementation of cellular neural networks, Real-Time Cellular Neural Network (CNN) Processor-v2 (RTCNNP-v2). RTCNNP-v2 is unique with its capabaility of processing full-HD video streams. Beside that, it is fastest CNN impelementation reported to date. First, mathematical models of CNN are explained in order to introduce full signal range model which is used at RTCNNP-v2 architecture. Then, templates are illustrated based on full signal range model (FSR). Results are given in tables and templates are classified according to compability of FSR. Finally, future work is discussed for the next generation of Real-Time Cellular Neural Network Processor.