| The field of superresolution has seen a tremendous growth in interest over the past decade. High resolution images are crucial in several applications including medical imaging and diagnosis, military surveillance, satellite and astronomical imaging, and remote sensing. Constraints due to factors such as technology, cost, size, weight, and quality prevent the use of sensors with the desired resolution in image capture devices and consequently, necessitate the design of superresolution algorithms to achieve the desired image resolution.; Wavelets have emerged as a powerful tool in signal processing and many other fields. Second generation wavelets were recently introduced and their flexibility and versatility has resulted in an ever-growing number of applications. They have been used in fields ranging from the solution of partial differential equations to mesh refinement and modeling in computer graphics. Two prominent properties of second generation wavelets, viz. the ability to handle irregular sampling structures and the adaptation to arbitrary boundaries, which are at the heart of image sequence superresolution, motivated this research on superresolution algorithms based on second generation wavelets. The developed techniques also achieve simultaneous noise filtering by thresholding the computed wavelet coefficients prior to reconstruction. Analysis leading to the selection of a threshold that yields an optimal trade-off between noise reduction and blur introduction clue to thresholding is subsequently presented. Since the choice of prediction neighborhood in second generation wavelet transforms influences reconstructed image quality, an adaptive neighborhood based on approximated gradients is proposed to enhance the quality of edges in the reconstructed images. Simulation results that demonstrate the superior performance of the developed techniques are also included.; In addition to noise, blur commonly affects the quality of the captured images/video. Typically, a deblurring module is employed following superresolution to counteract this degradation. In this research, an iterative blind deconvolution algorithm that generalizes directly to the multichannel framework is significantly enhanced by incorporating iterative estimation of the support of the blur. An expression for in-loop estimation of the asymmetry factor, which is used to compensate for the varying rates of convergence of image and blurring function estimates, is derived to replace the trial and error haled approach employed in the original algorithm. The efficacy of the proposed enhancements is evidenced by the presented simulation results. |
