In order to construct the Magnetic Resonance Imaging(MRI) database from online literature, MRI image recognition and brain MRI recognition are studied. In this paper, two complementary features, Pyramid Histogram of Orientated Gradients(PHOG) and Pyramid Histogram of Words(PHOW) are adopted to extract and describe the features of images. An improved Support Vector Machine(SVM) classifier based on feature fusion which combines spatial shape and local distribution information is proposed. Experimental result shows a significant improvement in the average accuracy of the fusion classifier as compared with classifiers only based on PHOG or PHOW. It provides a foundation of building a knowledge base system that can interpret MRI images in online articles.
It is difficult for model-unknown non-hyperbolic non-linear sequence to realize denoising or shadowing of non-linear time sequences. Aiming at the problem, based on the minimal description length criteria, this paper points out that the marginal error algorithm is not optimal to the discrete system, sometimes even inadaptable from the viewpoint of over-fitting and under-fitting. A modified algorithm named total error algorithm is proposed and analyzed. And a highly-stable-but-fast hybrid algorithm is developed which compensates both the slower convergence of gradient decent algorithm and the worse stability of Newton-Raphson algorithm. The machine precision is obtained for the noisy time sequences of model- unknown non-linear discrete system. Experimental results prove that the method is efficient to solve the denoising or shadowing problem of non-linear time sequences.
In liquid and smoke simulation using Euler model, the common Semi-Lagrangian Method(SLM) has numerical viscosity due to excessive averaging steps, and the surface extracting method will further smear out details of water surface. This paper proposes a stable and detail-preserving liquid and smoke simulation algorithm. It employs numerical method with higher numerical accuracy to solve advection term. The algorithm distributes massless implicit particles near the surface and in the fluid volume, using simple interpolating strategy and utilizing the bidirectional influence between particles and grid, achieving the target of guaranteeing numerical stability and reducing numerical viscosity. It combines with surface tracking method based on explicit surface to achieve rich visual effects. Experimental results demonstrate the algorithm has the capability of preserving details.
In order to solve the problem, i.e. finding out the signals and modules that are more vulnerable to environmental disturbance and should be assigned errors detecting and correcting mechanisms or fault tolerance techniques, this paper designs a method to analyze error propagation rules and an implement algorithm is put forward. Some reliability indices are introduced such as error propagating rate and error exposure rate. The error detecting and correcting mechanisms can be deployed on those more vulnerable signals and modules. Through a fault injection experiment, all indices are computed. At the same time, the experiment process proves the effectiveness and validity of algorithm.
This paper introduces an Edge-independent Time Evolving Graph(E-TEG) model to capture the evolution of the connectivity properties of Opportunistic Networks(OppNet). E-TEG model is presented through using discrete time Markovian model to deal with the time dependencies of consecutive time-step indexed network snapshots, and the dynamic of each possible edge is assumed to be an independent birth-death process. In addition, given the sequence data, the birth and the death probability of each edge are estimated through using Laplace’s rule of succession. It shows that an E-TEG eventually converges to an un-uniform random graph. E-TEG model is validated through CRAWDAD trace datasets by computing the fastest path of each pair of nodes in an instance of E-TEG.
In order to detect boundary points of clustering automatically and effectively, and to eliminate the impact of parameters on the results of the boundary detection, a new nonparametric boundary detection algorithm based on delaunay triangulation is presented. This algorithm calculates the boundary degree for each point in the generated delaunay triangulation without any parameters. According to the boundary degree’s threshold that is automatically calculated by k-means, dataset is divided into two parts: candidate set of boundary points and the set of non-boundary points. Based on the characteristics of the noise points, the noise points are removed from the candidate set of boundary points. It detects out boundary points of clustering. Experimental results show that the algorithm can identify boundary points in noisy datasets containing clustering of different shapes and sizes effectively and efficiently.