Research On Super Resolution Reconstruction Of Video And Image Sequences

With the rapid improvements of video and image processing technologies in recent years, the demand for high-quality video and image sequences grows fast. A high-quality image always contains further detailed information of targets, and it is of great value for analysis and post-process. But in some application areas, under limited optical elements, processors, channel bandwidths or storage capacities, the image resolution is always unable to meet our needs. Furthermore, it is impossible or hard to break the limitations. So, how to enhance the spatial resolution of video and image sequences under these limitations becomes a research hotspot.Super-resolution image reconstruction technique has been proved to be an efficient technique to solve the above problems. It fuses complementary information of several low resolution images by signal processing methods to get a high resolution image. It can enhance the spatial resolution of images effectively without any upgrade of current equipments. This technique provides us with an efficient approach to obtain high-quality videos and images subject to the constraints of optical devices, processors or communication channels. Thereby, it is worthy of notice both for academic studies and applications, and it permits widespread deployment.The dissertation investigates several key issues of super-resolution of video and image sequences including blur parameter estimation, block based motion estimation and regularization based image reconstruction, and has obtained many results.The main contributions and innovation points of the dissertation are as follows:I. A novel adaptive estimation algorithm of blur parameter for super resolution image reconstruction is proposed. The PSF (Point Spread Function) of optical device for a certain video sequence is always unavailable, but it must be preset for reconstruction process. By exploiting the mutual validation property among all the low resolution images, this algorithm adjusts and evaluates the blur parameter automatically, and the best parameter is acquired by training process. Simulation results show that the method can further preserve the detailed texture, and obtain high resolution image with better quality.II. A novel pyramidal motion estimation algorithm based on down-sampling of search points is proposed, and its IP architecture is designed under certain project requirement. We analyze the shortage of current pyramidal algorithms, and replace pixel down-sampling with searching-point down-sampling. By doing so, false selection of candidate match point at high levels is avoided. In VLSI implementation, each processing level has similar architecture, and hardware resources are saved obviously. The algorithm reduces the total computation load and occupancy rate of hardware resources sharply at the cost of little decline of image quality, and it meets the requirement of real-time video processing.III. A novel MAP super-resolution image reconstruction algorithm based on trilateral regularization is proposed. The new regularization function could better preserve slope and roof edge while keep global smoothing. This algorithm can adjust the parameters of regularization function automatically during processing. Simulation results on synthetic image and real video sequences confirm the effectiveness of this algorithm and demonstrate its superiority to other super resolution algorithms.IV. A novel MAP algorithm with embedded regularization is proposed based on the analysis and improvement of integration method of similarity term and regularization term in iterative solving function. By embedding regularization in similarity term, the rectify efficiency of regularization term is improved and the convergence rate is accelerated. Simulation results of synthetic images and real video confirm that, the convergence rate is faster and the reconstructed image quality is higher than other Tikhonov regularization based algorithms.