Investigation and analysis of optimization algorithms for multisensory image fusion

Abstract

The research work presented in this thesis is motivated by the need for maximizing performance in a multi-sensor image fusion system for enhancing the visualization of image data. Image fusion is the process of blending the most pertinent information from multiple source images for obtaining a comprehensive fused image, which contains rich and accurate information, making it suitable for further image processing tasks. Recent literature on multi-sensor image fusion indicates dependence of the fusion performance on the choice of the fusion rule and the algorithm for integration of source information. The existing image fusion techniques simply combines image data without analysis of the information content of the source images, which affects the spectral characteristics of the fused image. In this thesis, image fusion is formulated as an optimization problem using the multiresolution based image decomposition technique while swarm intelligence based optimization technique is used for effective combination of the information from multi-sensor images without any loss and spectral distortion. To start with, the proposed image fusion algorithm, Dual Tree Discrete Wavelet Transform (DTDWT) is applied for image decomposition and Particle Swarm Optimization (PSO) is used for obtaining the optimal weights so as to maximize the Entropy (E) and minimize Root Mean Square Error (RMSE) of the fused image. The reason to choose PSO is that it has faster convergence compared with other optimization algorithms. The robustness of proposed fusion algorithm is shown by evaluating the fused images with distorted input images by the addition of Gaussian white noise and Gaussian blur. newline