Multiframe Super Resolution Reconstruction Based On Texture And Detail Information

The high resolution image provides more detail information for almost every aspect of our daily life, such as video surveillance, scientific research, and military et al. However, the sensor measurement error, the noise, or the shooting environment would lead the image degraded such as the blurred edges, detail information lost. Multi-frame super resolution aims to obtain a higher resolution image by using a set of low resolution images in the same scene, which is a hot research point in image processing.One of the most important issues in multi-frame super resolution is how to preserve the edges, restore the textures by adding a prior constraint, as it is an ill-posed problem. However, the currently used prior model cannot restore the detail information as much as possible because the image is far more complex, which not only contains the edge regions but also contains texture regions or flat regions.Based on the above problems, starting with the maximum a posteriori method and variance Bayesian inference, focusing on prior model and observation model, a more efficiency and effective multi-frame super resolution reconstruction method have been proposed. The main content can be summarized as follows:Based on the maximum a posteriori method, which is the most popularly used method, an anisotropic fractional order adaptive model is proposed to not only preserve the edges but also restore the textures as much as possible. Then the proposed anisotropic fractional order adaptive model is applied into multi-frame super resolution as a prior model. Compared with the existing models, the proposed model can remove the noise and protect the edges adaptively according to the local features of the image.Inspired by the advantages of the TV norm and the H1 norm, an adaptively mixture norm named lmix norm is proposed and then is introduced it into multi-frame super resolution reconstruction as a prior model. The proposed 1mix model could assign the weights to the TV norm and the H1 norm according to the local gradient of the image. The proposed super resolution method is solved by using the Bayesian inference method in the hierarchical Bayesian framework because the a posteriori method lacks of convergence accuracy and have the shock phenomenon. The theoretical analysis and the experimental results show that the proposed method could reconstruct the image more efficiently. Compared with the existing combined model, the proposed model reduces human intervention and improves the reconstruction efficiency.The signals have been used in variance aspects of our modern life. An accuracy, effective and efficiency observation model is the basis of the signal processing. The signal degradation process is very complex and the currently used observation model is an idealized one, which haven't considered the information lost in degradation process. Therefore, in this paper, a novel observation model is proposed, which integrate the missing information into signal reconstruction. In this paper, we take super-resolution as an example to test the novel proposed observation model. Theoretical analysis and experimental results show that the proposed observation model could estimate more lost information.The image is reconstructed by the proposed the observation model and a novel adaptive non-local edge-preserving prior model, which could better preserve the detail information of an image while avoiding artifacts. The adaptive non-local edge-preserving prior model restores the pixel information in a larger region by using the non-local similarity. The experimental results show that the proposed method can reconstruct higher quality images in both quantitative term and perceptual effect.In summary, in this paper, in order to reconstruct the super resolution image with clear edges and rich texture details, the degradation process and different norms are deeply researched. Compared with the existing methods, the proposed methods could reconstruct the image with higher quality and more robustness, and efficiently reduce the reconstruction error.