Search

In recent years deep learning methods have shown great superiority in compressed video quality enhancement tasks. Existing methods generally take the raw video as the ground truth and extract practical information from consecutive frames containing various artifacts. However, they do not fully exploit the valid information of compressed and raw videos to guide the quality enhancement for compressed videos. In this paper, we propose a unique Valid Information Guidance scheme (VIG) to enhance the quality of compressed videos by mining valid information from both compressed videos and raw videos. Specifically, we propose an efficient framework, Compressed Redundancy Filtering (CRF) network, to balance speed and enhancement. After removing the redundancy by filtering the information, CRF can use the valid information of the compressed video to reconstruct the texture. Furthermore, we propose a progressive Truth Guidance Distillation (TGD) strategy, which does not need to design additional teacher models and distillation loss functions. By only using the ground truth as input to guide the model to aggregate the correct spatio-temporal correspondence across the raw frames, TGD can significantly improve the enhancement effect without increasing the extra training cost. Extensive experiments show that our method achieves the state-of-the-art performance of compressed video quality enhancement in terms of accuracy and efficiency.

According to the pathological characteristics of skin hypertrophic scar tissue, texture features on the edge direction, density distribution and geometric morphology are extracted from two-photon excited microscopic images of scar tissue and used to distinguish the normal tissue from scar tissue. Analysis algorithms of mean-shift are applied to analyze extracted texture features. Finally, a compositive formula is presented for quantitative analysis images. Twenty-five two-photon excited microscopic images originated from human hypertrophic scar tissue are statistically analyzed based on this formula and the inaccuracy rate is 4%. These results indicate that texture features and analytic methods proposed in this paper are effective on distinguishing the normal tissue from scar tissue and accurately fixing on their border.

Denoising biomedical images can be completed by a spatial averaging of nearby pixels. Neighborhood filters are usually preferred. In this paper, we perform an analysis of neighborhood filters where the size of the gray Euclidean distance is small. We prove that the weak edges and details are smoothed and the gray level transition tends to become mild filtering with these neighborhood filters. As a solution, we propose a gray-natural logarithm ratio bilateral neighborhood filter. The new gray-natural logarithm ratio bilateral neighborhood filter method can not only well restrain noise but also keep much more weak edges and details of a biomedical image, and preserve the original gray level transition of gray level images. Experimental results show that the effectiveness for biomedical image denoising with our method is better.

Biomedical image denoising algorithm based on gradient dependent energy functional often compromised the biomedical image features like textures or certain details. This paper proposes an iterative regularization denoising method based on OSV model for biomedical image denoising. By using iterative regularization, the oscillating patterns of texture and detail are added back to fit and compute the original OSV model,and the iterative behavior avoids overfull smoothing while denoising the features of textures and details to a certain extent. In addition, the iterative procedure is proposed in this paper, and the proposed algorithm also be proved the convergence property. Experimental results show that the proposed method can achieve a batter result in preserving not only the features of textures for biomedical image denoising but also the details for biomedical image.

Neighborhood filters are modern and powerful techniques for image denoising. In this paper, we perform an analysis for the normalization Factor of Neighborhood filters, where the size of the gray Euclidean distance is small. We prove that the weak edges and details are smoothed, and the gray level transition tends to become mild, when filtering with these Neighborhood filters. As a solution, a new gray-level value logarithm ratio range filter kernel is pointed out for the normalization Factor of Bilateral filter, a representative method of Neighborhood filters. The new gray-level value logarithm ratio Neighborhood filters method can not only restrain noise well but also keep much more weak edges and details of images, and preserve the original gray level or color transition for gray images or color images.

In the study of physics, many problems require to investigate the nonequilibrium ionization processes. For example, some material is irradiated by laser, then the plasma with low density and higher temperature may be produced. The radiation and the material are in nonequilibrium states. If the target is made of high Z elements, the electrons are ionized generally not completly. In order to study the absorption of laser energy, plasma dynamics and specially the transformation from laser into x-ray, one consider the ionizability and electron distribution with different occupation states. To describe such physical phenomena, the fluid dynamics systems, multi-group energy equations of photons, electrons, ions and equations for electron occupation probabilities are considered. There are nonlinear coupled ordinary and partial differential equations, which have much different time scales. It is a large scale scientific computing problem.