This paper proposes a method of modeling the reconfigurable core cell based on stochastic Petri net. For its structural properties, such as reachability, boundedness and safety, it presents the logical correctness verification. As to its performance properties, such as stable probability, transition probability and throughput rate, it presents the quantification analysis. This paper discusses the effects between dynamic reconfiguration time and the finished time of computing task under three different conditions. Analysis result shows that the shorter dynamic reconfigurable time is, the shorter task complete time is.
This paper proposes a method of multi-view face features localization. The face is located by AdaBoost detector and the search ranges of the face features are determined. The candidate eye, nose and mouth regions are found by the improved Support Vector Machine(SVM) detectors trained by large scale multi-view face features examples, which use the brow-eye and nose-mouth features. The candidate eye, nose and mouth regions are filtered and merged to refine their location on the multi-view face. Experimental results show that the method has very good accuracy and robustness to the face features localization with various face post and expression in the complex background.
Aiming at the shortcoming of the compressed edge fragment sampling algorithm, a new method is proposed, called Overlapping Hash Fragment(OHF) Probabilistic Packet Marking(PPM) method. The new method reduces the computational complexity during reconstruction by constructing 4 bits hash relationship between the adjacent IP fragments without increasing the marking amounts. And the new method improves the false alarm rate. Simulation experiment in the NS2 show the validity of the method.
To solve automatic predicate-verb choosing for argument, this paper gives semantics preference method based on Minumum Description Length(MDL) and Latent Semantic Clustering(LSC). MDL is used to calculate of each verb-noun pair. The probabilities of a verb preferring for a noun P(v,n) is computed based on LSC model and EM is used to evaluate the parameters. For the same verb-noun pair, the sum of and P(v,n) is considered to represent the association between the verb and the noun. Experiments show the F1 reaches 85.26%, and it is better than MDL or SCL methods.
Considering the outer classes and inferior problem in Kernel Maximum Scatter Difference(KMSD) method, a new method of face recognition based on Kernel Principal Component Analysis(KPCA) and Fuzzy Maximum Scatter Difference(FMSD) is developed. The KPCA can be benefit to develop the nonlinear structures features in faces. Selecting the eigenvectors that between-class scatter is greater than within-class scatter after projection as optimal projection axis. Distribution information of samples is represented with fuzzy membership degree in the FMSD. It uses the nearest neighbor classifier for face recognition. Experimental results on ORL and YALE face databases show the KFMSD is better than KMSD method.