Mathematical Modeling and Simulation of Digital Image Processing Problems

Abstract

The objective of this thesis is to model and simulate different image processing problems. By identifying the crucial parameters in existing image processing techniques and in artificial neural network models, algorithms for image deblurring, image data compression and pattern recognition are modeled and simulated with better accuracies. newlineA general problem faced in captured image is due to the blur and it is also observed that some amount of blurring inevitably occurs in recording of digital images. Six deblurring algorithms have been proposed in this thesis. These algorithms are proposed alongwith the safety window. The performance of these algorithms for background removal is tested quantitatively by calculating SNRout. Pattern recognition is also an important task in image processing applications. Today we have devices for easier interface between human and computer. But it is necessary for more realistic human computer cooperation as computer usage reaches to a bigger mass. Character recognition is one such option. In this thesis, the perceptron learning rule is used for pattern recognition of noisy characters. In comparison to multilayer perceptron neural network this rule reduces the complexity of the network. newlineIn real time applications when environment is constantly changing, an autonomous learning system is needed, which maintains its plasticity or adaptability to the significant incidents and can also simultaneously stabilizes itself to the insignificant incidents. Adaptive resonance theory, ART provides a solution to this. The thesis also proposes the pattern recognition through ART1 neural network. In digital image processing applications for the sake of saving disk memory space and reducing transmission rate, images need to be compressed. For JPEG and JPEG 2000 images there are many techniques and standards for data compression. Huffman codes and B codes are normally used in the entropy coding phase. These compression techniques have been compared in this thesis for entropy and average word length. newline