High resolution image reconstruction from multiple degraded images acquired with multisensors

While a high resolution image is strongly desired in many applications including television broadcasting, teleconferencing, commercial electronics and defense, physical restrictions on the size of current sensing elements make it difficult for a single sensor to capture the desired high resolution image. Therefore, a high resolution image needs to be reconstructed from multiple undersampled images which are obtained by using multiple identical image sensors shifted from each other by subpixel displacements. Since image reconstruction problem from a multisensor array system can be modeled as image restoration, various aspects of image restoration problem such as boundary condition, regularization and blur estimation should be considered. Also, an efficient preconditioner should be employed to process high resolution image reconstruction within given time. While high resolution image restoration from a multisensor array has been a subject of active research, regularization and boundary condition have been less emphasized.; With Neumann boundary condition imposed, the boundary error is analyzed and a simple modification of the observed image is suggested to reduce boundary error. Neumann boundary condition and the suggested simple modification scheme is superior to the conventional periodic boundary condition.; The L-curve method is known to be a robust algorithm for choosing proper regularization parameter in ill-posed problem including image restoration. When the subpixel displacement errors exist in a multisensor array system, the resulting system matrix is difficult to invert, which makes computation of L-curve points expensive. Using Lanczos bidiagonalization algorithm, an efficient method to compute L-curve points for the case of space-variant regularization is provided.; Most works on high resolution image restoration from multisensor array are based on the assumption that the subpixel displacement errors are known a priori. Adopting space variant regularization and total least squares algorithm, precise estimation of sub-pixel displacement errors and improved image quality are accomplished. The efficient preconditioner for the resulting system matrix is considered. Besides monochrome image, high resolution color image restoration from data acquired through a multisensor array is also considered.