| Since entering the digital imaging era, driven by actual demands, people are never stopping the pursuit for the higher resolution images. If using image super-resolution reconstruction technology, people can reconstruct a higher resolution image from low-resolution image sequence with complementary information by using some algorithms under the situation without changing the existing hardware. The reconstructive results not only can improve the subjective visual effect, but also are significant for the follow-up image processing, such as image segmentation, feature extraction, identification and so on.In this paper, optical still image sequence and video streaming are used as main object of study. We focus on the key technologies of multi-frame image SR reconstruction. Main research works are shown as follow. (1) Systematically summarize, analyze and compare the main SR theories and methods which formed in the development of the research field. And its mathematical basis are analyzed and discussed. To solve the problem that abusing the reconstructive models and unstable reconstructive results which is caused by the inadequate analysis about the application conditions and application range of the models, two different models are established that one is based on warping and the other is based on the motion blur, and proposed a reconstruction model and framework based on different degraded models.(2) For multi-frame motion blur degraded images, a theory and algorithm is proposed for multi-frame estimation for motion-blur parameters and multi-frame nonlinear reconstruction. Under this framework, for solving the problem that parameter estimation for any direction of motion-blur, an algorithm is proposed that using two-dimensional cepstrum analysis to estimate the parameters of PSF. In order to reduce wraparound artifact, we proposed the basic idea of how to suppress wraparound artifact and specific algorithm to do it based on detailed analysis of the relationship between wraparound artifact and various factors producing it. Finally, for improving the resolution of motion-blur images, a blind multi-frame non-linear reconstruction algorithm and framework is proposed.(3) For multi-frames (video streaming) that not main degraded by motion blur, the theory and algorithm of blind multi-frame SR reconstruction based on Gauss-POCS model is proposed. Several key technologies have been studied as follow. For obtained high-precision initial estimates, a bilateral filter interpolation algorithm is proposed which can maintain edge features well. For motion estimation problem, we proposed multi-frame subpixel motion-estimation algorithms which use spatio-temporal gradient iterative and combine global model and local model. For solving reconstruction problem, we use POCS as the theoretical basis to establish a variety of constraints and the corresponding projection operations. Data consistency constraints have been improved. A post-suppressed algorithm for wraparound artifact is proposed. Multi-frame joint denoising method is introduced. Finally, we establish a blind SR reconstruction algorithm based on Gauss-POCS model to obtain the image that subjective visual effects and objective assessment are greatly improved.(4) Establishment of efficient evaluation system for SR algorithm has always been a difficult and challenging research area in SR research. For research this problem, firstly the existing digital image quality evaluation methods are sorted out. And a variety of indicators are used to evaluate Gaussian blur and Gaussian noise which are two typical degraded types in SR research field. Then, current evaluation situation of SR reconstruction has been analyzed and existing problems has been identified. We proposed some ideas and a few basic principles that establish evaluation system for SR reconstruction based on above analysis. Finally, an effective post-suppressed algorithm and assessment index for wraparound artifact which is produced by SR reconstruction is proposed. |
PDF Full Text Request