Color to Gray: Visual Cue Preservation

Both commercial and scientific applications often need to transform color images into gray-scale images, e. g., to reduce the publication cost in printing color images or to help color blind people see visual cues of color images. However, conventional color to gray algorithms are not ready for practical applications because they encounter the following problems: 1) Visual cues are not well defined so it is unclear how to preserve important cues in the transformed gray-scale images; 2) some algorithms have extremely high time cost for computation; and 3) some require human-computer interactions to have a reasonable transformation. To solve or at least reduce these problems, we propose a new algorithm based on a probabilistic graphical model with the assumption that the image is defined over a Markov random field. Thus, color to gray procedure can be regarded as a labeling process to preserve the newly well-defined visual cues of a color image in the transformed gray-scale image. Visual cues are measurements that can be extracted from a color image by a perceiver. They indicate the state of some properties of the image that the perceiver is interested in perceiving. Different people may perceive different cues from the same color image and three cues are defined in this paper, namely, color spatial consistency, image structure information, and color channel perception priority. We cast color to gray as a visual cue preservation procedure based on a probabilistic graphical model and optimize the model based on an integral minimization problem. We apply the new algorithm to both natural color images and artificial pictures, and demonstrate that the proposed approach outperforms representative conventional algorithms in terms of effectiveness and efficiency. In addition, it requires no human-computer interactions.

Smoothing LDA Model for Text Categorization

Abstract. Latent Dirichlet Allocation (LDA) is a document level language model. In general, LDA employ the symmetry Dirichlet distribution as prior of the topic-words’ distributions to implement model smoothing. In this paper, we propose a data-driven smoothing strategy in which probability mass is allocated from smoothing-data to latent variables by the intrinsic inference procedure of LDA. In such a way, the arbitrariness of choosing latent variables'priors for the multi-level graphical model is overcome. Following this data-driven strategy,two concrete methods, Laplacian smoothing and Jelinek-Mercer smoothing, are employed to LDA model. Evaluations on different text categorization collections show data-driven smoothing can significantly improve the performance in balanced and unbalanced corpora.

三峡库区滑坡的时空分布、成因及其预报模式

Landslides are widely distributed along the main stream banks of the Three Gorges Reservoir area. Especially with the acceleration of the human economic activities in the recent 30 years, the occurrence of landslide hazards in the local area trends to be more serious. Because of the special geological, topographic and climatic conditions of the Three Gorges areas, many Paleo-landslides are found along the gentle slope terrain of the population relocation sites. Under the natural condition, the Paleo-landslides usually keep stable. The Paleo-landslides might revive while they are influenced under the strong rainfall, water storage and migration engineering disturbance. Therefore, the prediction and prevention of landslide hazards have become the important problem involving with the safety of migration engineering of the Three Gorges Reservoir area.The past research on the landslides of the Three Gorges area is mainly concentrated on the stability analysis of individual landslide, and importance was little attached to the knowledge on the geological environment background of the formation of regional landslides. So, the relationship between distribution and evolution of landslides and globe dynamic processes was very scarce in the past research. With further study, it becomes difficult to explain the reasons for the magnitude and frequency of major geological hazards in terms of single endogenic or exogenic processes. It is possible to resolve the causes of major landslides in the Three Gorges area through the systematic research of regional tectonics and river evolution history.In present paper, based on the view of coupling of earth's endogenic and exogenic processes, the author researches the temporal and spacial distribution and formation evolution of major landslides(Volume^lOOX 104m3) in the Three Gorges Reservoir area through integration of first-hand sources statistics, .geological evolution history, isotope dating and numerical simulation method etc. And considering the main formation factors of landslides (topography, geology and rainfall condition), the author discusses the occurrence probability and prediction model of rainfall induced landslides.The distribution and magnitude of Paleo-landslides in the Three Gorges area is mainly controlled by lithology, geological structure, bank slope shape and geostress field etc. The major Paleo-landslides are concentrated on the periods 2.7-15.0 X 104aB.R, which conrresponds to the warm and wettest Paleoclimate stages. In the same time, the Three Gorges area experiences with the quickest crust uplift phase since 15.0X 104aB.P. It is indicated that the dynamic factor of polyphase major Paleo-landslides is the coupling processes of neotectonic movement and Quaternary climate changes. Based on the numerical simulation results of the formation evolution of Baota landslide, the quick crust uplift makes the deep river incision and the geostress relief causes the rock body of banks flexible. Under the strong rainfall condition, the pore-water pressure resulted from rain penetration and high flood level can have the shear strength of weak structural plane decrease to a great degree. Therefore, the bank slope is easy to slide at the slope bottom where shear stress concentrates. Finally, it forms the composite draught-traction type landslide of dip stratified rocks.The susceptibility idea for the rainfall induced landslide is put forward in this paper and the degree of susceptibility is graded in terms of the topography and geological conditions of landslides. Base on the integration with geological environment factors and rainfall condition, the author gives a new probabilistic prediction model for rainfall induced landslides. As an example from Chongqing City of the Three Gorges area, selecting the 5 factors of topography, lithology combination, slope shape, rock structure and hydrogeology and 21 kinds of status as prediction variables, the susceptibility zonation is carried out by information methods. The prediction criterion of landslides is established by two factors: the maximum 24 hour rainfall and the antecedent effective precipitation of 15 days. The new prediction model is possible to actualize the real-time regional landslide prediction and improve accuracy of landslide forecast.

A probabilistic model for fatigue crack propagation analysis

A simple probabilistic model for predicting crack growth behavior under random loading is presented. In the model, the parameters c and m in the Paris-Erdogan Equation are taken as random variables, and their stochastic characteristic values are obtained through fatigue crack propagation tests on an offshore structural steel under constant amplitude loading. Furthermore, by using the Monte Carlo simulation technique, the fatigue crack propagation life to reach a given crack length is predicted. The tests are conducted to verify the applicability of the theoretical prediction of the fatigue crack propagation.

A nonlinear model for assessing multiple probabilistic risks: A case study in South five-island of Changdao National Nature Reserve in China

Several methods for estimating the potential impacts caused by multiple probabilistic risks have been suggested. These existing methods mostly rely on the weight sum algorithm to address the need for integrated risk assessment. This paper develops a nonlinear model to perform such an assessment. The joint probability algorithm has been applied to the model development. An application of the developed model in South five-island of Changdao National Nature Reserve, China, combining remote sensing data and a GIS technique, provides a reasonable risk assessment. Based on the case study, we discuss the feasibility of the model. We propose that the model has the potential for use in identifying the regional primary stressor, investigating the most vulnerable habitat, and assessing the integrated impact of multiple stressors. (C) 2006 Elsevier Ltd. All rights reserved.

Probabilistic Modeling of Rosette Formation

英文摘要: Rosetting, or forming a cell aggregate between a single target nucleated cell and a number of red blood cells (RBCs), is a simple assay for cell adhesion-mediated by specific receptor-ligand interaction. For example, rosette formation between sheep RBC and human lymphocytes has been used to differentiate T cells from B cells. Rosetting assay is commonly used to determine the interaction of Fc gamma-receptors (Fc gamma R) expressed on inflammatory cells and IgG-coated on RBCs. Despite its wide use in measuring cell adhesion, the biophysical parameters of rosette formation have not been well characterized. Here we developed a probabilistic model to describe the distribution of rosette sizes, which is Poissonian. The average rosette size is predicted to be proportional to the apparent two-dimensional binding affinity of the interacting receptor-ligand pair and their site densities. The model has been supported by experiments of rosettes mediated by four molecular interactions: Fc gamma RIII interacting with IgG, T cell receptor and coreceptor CD8 interacting with antigen peptide presented by major histocompatibility molecule, P-selectin interacting with P-selectin glycoprotein ligand 1 (PSGL-1), and L-selectin interacting with PSGL-1. The latter two are structurally similar and are different from the former two. Fitting the model to data enabled us to evaluate the apparent effective two-dimensional binding affinity of the interacting molecular pairs: 7.19x10(-5) mu m(4) for Fc gamma RIII-IgG interaction, 4.66x10(-3) mu m(4) for P-selectin-PSGL-1 interaction, and 0.94x10(-3) mu m(4) for L-selectin-PSGL-1 interaction. These results elucidate the biophysical mechanism of rosette formation and enable it to become a semiquantitative assay that relates the rosette size to the effective affinity for receptor-ligand binding.

Probabilistic modelling of soil moisture dynamics of irrigated cropland in the North China Plain

A probabilistic soil moisture dynamic model is used to estimate the soil moisture probability distribution and plant water stress of irrigated cropland in the North China Plain. Soil moisture and meteorological data during the period of 1998 to 2003 were obtained from an irrigated cropland ecosystem with winter wheat and maize in the North China Plain to test the probabilistic soil moisture dynamic model. Results showed that the model was able to capture the soil moisture dynamics and estimate long-term water balance reasonably well when little soil water deficit existed. The prediction of mean plant water stress during winter wheat and maize growing season quantified the suitability of the wheat-maize rotation to the soil and climate environmental conditions in North China Plain under the impact of irrigation. Under the impact of precipitation fluctuations, there is no significant bimodality of the average soil moisture probability density function.

Supervised Gaussian Process Latent Variable Model for Dimensionality Reduction

The Gaussian process latent variable model (GP-LVM) has been identified to be an effective probabilistic approach for dimensionality reduction because it can obtain a low-dimensional manifold of a data set in an unsupervised fashion. Consequently, the GP-LVM is insufficient for supervised learning tasks (e. g., classification and regression) because it ignores the class label information for dimensionality reduction. In this paper, a supervised GP-LVM is developed for supervised learning tasks, and the maximum a posteriori algorithm is introduced to estimate positions of all samples in the latent variable space. We present experimental evidences suggesting that the supervised GP-LVM is able to use the class label information effectively, and thus, it outperforms the GP-LVM and the discriminative extension of the GP-LVM consistently. The comparison with some supervised classification methods, such as Gaussian process classification and support vector machines, is also given to illustrate the advantage of the proposed method.