About: In this paper, we propose an improved principal component analysis based on maximum entropy (MaxEnt) preservation, called MaxEntPCA, which is derived from a Parzen window estimation of Renyi’s quadratic entropy. Instead of minimizing the reconstruction error either based on L2norm or L1norm, the MaxEntPCA attempts to preserve as much as possible the uncertainty information of the data measured by entropy. The optimal solution of MaxEntPCA consists of the eigenvectors of a Laplacian probability matrix corresponding to the MaxEnt distribution. MaxEntPCA (1) is rotation invariant, (2) is free from any distribution assumption, and (3) is robust to outliers. Extensive experiments on realworld datasets demonstrate the effectiveness of the proposed linear method as compared to other related robust PCA methods. Changes:Initial Announcement on mloss.org.

About: MetropolisHastings alogrithm is a Markov chain Monte Carlo method for obtaining a sequence of random samples from a probability distribution for which direct sampling is difficult. Thi sequence can be used to approximate the distribution. Changes:Initial Announcement on mloss.org.

About: This code is developed based on Uriel Roque's active set algorithm for the linear least squares problem with nonnegative variables in: Portugal, L.; Judice, J.; and Vicente, L. 1994. A comparison of block pivoting and interiorpoint algorithms for linear least squares problems with nonnegative variables. Mathematics of Computation 63(208):625643.Ran He, WeiShi Zheng and Baogang Hu, "Maximum Correntropy Criterion for Robust Face Recognition," IEEE TPAMI, in press, 2011. Changes:Initial Announcement on mloss.org.

About: Urheen is a toolkit for Chinese word segmentation, Chinese pos tagging, English tokenize, and English pos tagging. The Chinese word segmentation and pos tagging modules are trained with the Chinese Tree Bank 7.0. The English pos tagging module is trained with the WSJ English treebank(0223). Changes:Initial Announcement on mloss.org.

About: OpenPRNBEM is an C++ implementation of Naive Bayes Classifier, which is a wellknown generative classification algorithm for the application such as text classification. The Naive Bayes algorithm requires the probabilistic distribution to be discrete. OpenPRNBEM uses the multinomial event model for representation. The maximum likelihood estimate is used for supervised learning, and the expectationmaximization estimate is used for semisupervised and unsupervised learning. Changes:Initial Announcement on mloss.org.

About: This is a class to calculate histogram of LBP (local binary patterns) from an input image, histograms of LBPTOP (local binary patterns on three orthogonal planes) from an image sequence, histogram of the rotation invariant VLBP (volume local binary patterns) or uniform rotation invariant VLBP from an image sequence. Changes:Initial Announcement on mloss.org.

About: This program implements a novel robust sparse representation method, called the twostage sparse representation (TSR), for robust recognition on a largescale database. Based on the divide and conquer strategy, TSR divides the procedure of robust recognition into outlier detection stage and recognition stage. The extensive numerical experiments on several public databases demonstrate that the proposed TSR approach generally obtains better classification accuracy than the stateoftheart Sparse Representation Classification (SRC). At the same time, by using the TSR, a significant reduction of computational cost is reached by over fifty times in comparison with the SRC, which enables the TSR to be deployed more suitably for largescale dataset. Changes:Initial Announcement on mloss.org.

About: This is a implementation of the classic P3P(Perspective 3Points) algorithm problem solution in the Ransac paper "M. A. Fischler, R. C. Bolles. Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography. Comm. of the ACM, Vol 24, pp 381395, 1981.". The algorithm gives the four probable solutions of the P3P problem in about 0.1ms, and can be used as input of the consequent RANSAC step. The codes needs the numerics library VNL which is a part of the widely used computer vision library VXL. One can download & install it from http://vxl.sourceforge.net/. Changes:Initial Announcement on mloss.org.

About: It's a C++ program for symmetric matrix diagonalization, inversion and principal component anlaysis(PCA). The matrix diagonalization function can also be applied to the computation of singular value decomposition (SVD), Fisher linear discriminant analysis (FLDA) and kernel PCA (KPCA) if forming the symmetric matrix appropriately. Changes:Initial Announcement on mloss.org.

About: This program is a C++ implementation of Linear Discriminant Function Classifier. Discriminant functions such as perceptron criterion, cross entropy (CE) criterion, and least mean square (LMS) criterion (all for multiclass classification problems) are supported in it. The program uses a sparsedata structure to represent the feature vector to seek higher computational speed. Some other techniques such as online updating, weights averaging, gaussian prior regularization are also supported. Changes:Initial Announcement on mloss.org.

About: This program is a C++ implementation of Naive Bayes Classifier, which is a wellknown generative classification algorithm for the application such as text classification. The Naive Bayes algorithm requires the probabilistic distribution to be discrete. The program uses the multinomial event model for representation, the maximum likelihood estimate with a Laplace smoothing technique for learning parameters. A sparsedata structure is defined to represent the feature vector in the program to seek higher computational speed. Changes:Initial Announcement on mloss.org.

About: A simple and clear OpenCV based extended Kalman filter(EKF) abstract class implementation,absolutely following standard EKF equations. Special thanks to the open source project of KFilter1.3. It is easy to inherit it to implement a variable state and measurement EKF for computer vision and INS usages. Changes:Initial Announcement on mloss.org.

About: Supervised Latent Semantic Indexing(SLSI) is an supervised feature transformation method. The algorithms in this package are based on the iterative algorithm of Latent Semantic Indexing. Changes:Initial Announcement on mloss.org.

About: This program is used to extract SIFT points from an image. Changes:Initial Announcement on mloss.org.

About: LDPar is an efficient datadriven dependency parser. You can train your own parsing model on treebank data and parse new data using the induced model. Changes:Initial Announcement on mloss.org.

About: This program is used to extract HOG(histograms of oriented gradients) features from images. The integral histogram is used for fast histogram extraction. Both APIs and binary utility are provided. Changes:Initial Announcement on mloss.org.

About: This program is used to find point matches between two images. The procedure can be divided into two parts: 1) use SIFT matching algorithm to find sparse point matches between two images. 2) use "quasidense propagation" algorithm to get "quasidense" point matches. Changes:Initial Announcement on mloss.org.

About: Hofmann, T. 1999. Probabilistic latent semantic indexing. In Proceedings of the 22nd ACMSIGIR International Conference on Research and Development in Information Retrieval (Berkeley,Calif.), ACM, New York, 50–57. Changes:Initial Announcement on mloss.org.

About: MeanShift (MS) is a powerful nonparametric clustering method. Although good accuracy can be achieved, its computational cost is particularly expensive even on moderate data sets. For the purpose of algorithm speedup, an agglomerative MS clustering method called AggloMS was developed, along with its modeseeking ability and convergence property analysis. The method is built upon an iterative query set compression mechanism which is motivated by the quadratic bounding optimization nature of MS. The whole framework can be efficiently implemented in linear running time complexity. Changes:Initial Announcement on mloss.org.

About: The toolbox is to calculate normalized information measures from a given m by (m+1) confusion matrix for objective evaluations of an abstaining classifier. It includes total 24 normalized information measures based on three groups of definitions, that is, mutual information, information divergence, and cross entropy. Changes:Initial Announcement on mloss.org.
