Reduced Quality Of The Image Change Db Ye Sichao Distinguish With Hierarchical Adaptive Segmentation Algorithm

The image is a very important visual source for machine vision system. However, in the image acquisition and imaging system, there usually exists various degradations including motion warping, system blurring and noise effect. These degradations will not only cause the visual quality decline, but also affect the usage of the valid image information. For instance, in the case of performing segmentation of degradation image, the accuracy of segmentation has a serious decline, and thus affects the target detection, recognition, and understanding. Therefore preprocessing and segmentation of degradation image is an important scientific problem in image processing and computer vision. This article mainly research layered Bayesian super-resolution pretreatment of degradation image and layered adaptive image segmentation and expand research about the relevant algorithm.Firstly, using hierarchical Bayesian theory, we design and implemented a variational Bayesian super-resolution method based on horizontal and vertical gradient L1sparsity prior. Through the modeling of image degradation, image prior and hyper-parameters prior, we establish a multi-images variational super-resolution model, and we then implemented two kinds of image super-resolution corresponding blind and non-blind mode. Furthermore, using the YUV color space, the algorithms are extended to deal with the color images. Experimental results show that our method can obtain a good estimation for the registration parameters among multiple images and achieve a good super-resolved image.Secondly, we study a hierarchical image segmentation method mechanism based on algebraic multigrid. On this basis, though the establishment of a hierarchical graph model, we give a new top-down layered adaptive segmentation method of image. Experimental results verify that the algorithm can achieve automatic segmentation of multi-scale and in case of without interaction information, our segmentation can achieve a segmentation result with a good accuracy, closing with the segmentation performance of iterative graph cut which include interaction information.Finally, we design and implement a system of super-resolution and segmentation for degradation image. On the base of performing blind super-resolution pretreatment of degradation image, our system can perform segmentation of interesting object in degradation image and can achieve a good accuracy of segmentation. Our system can perform super-resolution construction in non-blind and blind two ways. The degradation image can be a grayscale image or a color image, and there are two kinds of super-resolution method for the color image. The system performs segmentation in automatic manner, and human-computer interaction is not required to be able to get the multilayer segmentation results.