Occurrence Matrix Research Articles

On the Influence of Latent Semantic Analysis Parameterization for Bug Localization

The bug localization problem has benefited from modern information retrieval techniques, such as Latent Semantic Analysis. There are many factors that influence the quality of results of this approach, such as, stop-words, term-document matrix transformations, dimensionality reduction and filtering criteria of the corpus. In this paper, we study the effect of different combinations for these factors on the impact of the accuracy of the query results in the proposed technique for bug localization. Bugs of three real-world software systems were analyzed with different combinations of input parameters for the LSA technique. Our results suggest that the term-document matrix transformations and filtering criteria of the corpus have major influence in the quality of the result and that the combination of adequate individual parameter values does not necessarily produce the best combination. Furthermore, some general guidance for parameterization of the LSA technique for bug localization could also be suggested from the observed results.

Extraction of color image texture feature based on gray-level co-occurrence matrix

Texture plays a very important role in image retrieval and classification,and texture feature extraction has been a research hotspot. Most present existing texture extraction algorithms can be only used to calculate texture features of gray image. Texture extraction algorithm for color image is very few. Referring to the analytical method of gray level co- occurrence matrix( GLCM),the authors analyzed the influence law of parameters( direction, distance,grayscale,window size) on GLCM texture features of color image. A color image texture feature extraction method( color GLCM,CGLCM) based on GLCM was realized. Through analyzing the influence law of these parameters on four texture features( ASM( angular second moment),Entropy,Contrast,Correlation),a proper parameter value range was given and the CGLCM method was optimized. The results of comparing CGLCM method with GLCM method show that the four texture features calculated with CGLCM method have better robustness and identification capability. These results can provide reference for image retrieval and classification based on texture information.

Combining Position Weight Matrices and Document-Term Matrix for Efficient Extraction of Associations of Methylated Genes and Diseases from Free Text

BackgroundIn a number of diseases, certain genes are reported to be strongly methylated and thus can serve as diagnostic markers in many cases. Scientific literature in digital form is an important source of information about methylated genes implicated in particular diseases. The large volume of the electronic text makes it difficult and impractical to search for this information manually.MethodologyWe developed a novel text mining methodology based on a new concept of position weight matrices (PWMs) for text representation and feature generation. We applied PWMs in conjunction with the document-term matrix to extract with high accuracy associations between methylated genes and diseases from free text. The performance results are based on large manually-classified data. Additionally, we developed a web-tool, DEMGD, which automates extraction of these associations from free text. DEMGD presents the extracted associations in summary tables and full reports in addition to evidence tagging of text with respect to genes, diseases and methylation words. The methodology we developed in this study can be applied to similar association extraction problems from free text.ConclusionThe new methodology developed in this study allows for efficient identification of associations between concepts. Our method applied to methylated genes in different diseases is implemented as a Web-tool, DEMGD, which is freely available at http://www.cbrc.kaust.edu.sa/demgd/. The data is available for online browsing and download.

SEMANTIC SEARCH OF SERVICES

This paper addresses semantic search of Web services using natural language processing. First we survey various existing approaches, focusing on the fact that the expensive costs of current semantic annotation frameworks result in limited use of semantic search for large scale applications. We then propose a service search framework based on the vector space model to combine the traditional frequency weighted term-document matrix, the syntactical information extracted from a lexical database and a dependency grammar parser. In particular, instead of using terms as the rows in a term-document matrix, we propose using synsets from WordNet to distinguish different meanings of a word under different contexts as well as clustering different words with similar meanings. Also based on the characteristics of Web services descriptions, we propose an approach to identifying semantically important terms to adjust weightings. Our experiments show that our approach achieves its goal well.

Visualizing of the structure of subject trends in Persian articles published during 2008–2012 in information organization domain

This research investigates structure of the subject trends in the Persian articles published during the period of 2008–2012 about “information organization” in Iranian journals and creates a semantic map for this domain. This structure has been considered in two sections: First, the weight of each subject term used in articles was indicated, and then the relationships amounts of terms were measured. Content analysis and weighting were used in the first section. The Vector Space Model formula was used in order to weight terms. In the second section, Co-term analysis was used, that is, the number of co-occurrence of any two terms in the same article. Then the term/document occurrence matrix of these terms was created and at last the relations significant between terms were measured by Pearson Correlation Coefficient. To illustrate the structure of information organization domain, Ucinet software was used. Findings show that high-weight categories in Persian articles are: “cataloging”, “information retrieval”, “information systems” and “authority files”. Furthermore through weighting, determined that in trends rate aspect, the terms “ontologies”, “cataloging”, “evaluation”, “Dewey Decimal Classification”, “fields”, “libraries and information centers” and “descriptors”, “Functional Requirements for Bibliographic Records”, “metadata elements” and “National library and archives of I.R. of Iran” are 10 highest-weight terms. Research findings about the relationships of terms show attention to “cataloging” by “IFLA” and “National library and archives of Iran” and critical approach to this subject. Some other findings about this show: “indexing” are considered by cataloging and thesauri, “metadata” are considered in web-based and information retrieval points of view, “ontologies” term is considered in approach of basic concepts, tools and relevance, and attention to some organizations such as “IFLA” and “National library and archives organization of Iran” besides “users” and their information needs. In terms of weighting and relationships between the subject terms in Persian articles, “cataloging” is in the center of information organization domain. The semantic map of information organization in Persian articles illustrates three region of subject area including: 1. Cataloging, 2. Ontologies, thesauri and information retrieval, and 3. Metadata and information retrieval.

Combining fractal and deterministic walkers for texture analysis and classification

In this paper, we present a novel texture analysis method based on deterministic partially self-avoiding walks and fractal dimension theory. After finding the attractors of the image (set of pixels) using deterministic partially self-avoiding walks, they are dilated in direction to the whole image by adding pixels according to their relevance. The relevance of each pixel is calculated as the shortest path between the pixel and the pixels that belongs to the attractors. The proposed texture analysis method is demonstrated to outperform popular and state-of-the-art methods (e.g. Fourier descriptors, occurrence matrix, Gabor filter and local binary patterns) as well as deterministic tourist walk method and recent fractal methods using well-known texture image datasets.

BRAIN TUMOR TISSUE CATEGORIZATION IN 3D MAGNETIC RESONANCE IMAGES USING IMPROVED PSO FOR EXTREME LEARNING MACHINE

Magnetic Resonance Imaging (MRI) technique is one of the most useful diagnostic tools for human soft tissue analysis. Moreover, the brain anatomy features and internal tissue architecture of brain tumor are a complex task in case of 3-D anatomy. The additional spatial relationship in transverse, longitudinal planes and the coronal plane information has been proved to be helpful for clinical applications. This study extends the computation of gray level co- occurrence matrix (GLCM) and Run length matrix (RLM) to a three- dimensional form for feature extraction. The sub-selection of rich optimal bank of features to model a classifler is achieved with custom Genetic Algorithm design. An improved Extreme Learning Machine (ELM) classifler algorithm is explored, for training single hidden layer artiflcial neural network, integrating an enhanced swarm-based method in optimization of the best parameters (input-weights, bias, norm and hidden neurons), enhancing generalization and conditioning of the algorithm. The method is modeled for automatic brain tissue and pathological tumor classiflcation and segmentation of 3D MRI tumor images. The method proposed demonstrates good generalization capability from the best individuals obtained in the learning phase to handle sparse image data on publically available benchmark dataset and real time data sets.

Content Based Image Retrieval Using Low Level Features of Automatically Extracted Regions of Interest

This paper proposes a region based image retrieval system using the local colour and texture features of image sub regions. The regions of interest (ROI) are roughly identified by segmenting the image into fixed partitions, finding the edge map and applying morphological dilation. The colour and texture features of the ROIs are computed from the histograms of the quantized HSV colour space and Gray Level co- occurrence matrix (GLCM) respectively. Each ROI of the query image is compared with same number of ROIs of the target image that are arranged in the descending order of white pixel density in the regions, using Euclidean distance measure for similarity computation. Preliminary experimental results show that the proposed method provides better retrieving result than retrieval using some of the existing methods. 

Dominant Color and Texture Approached for Content Based Video Images Retrieval

Content-based retrieval allows finding information by searching its content rather than its attributes. Content-based search and retrieval of video data becomes a challenging and important problem. Every year video content is growing in volume and there are different techniques available to capture, compress, display, store and transmit video while editing and manipulating video based on their content is still a non-trivial activity. Recent advances in multimedia technologies allow the capture and storage of video data with relatively inexpensive computers. However, without appropriate search techniques all these data are hardly usable. Today research is focused on video retrieval.Moreover, content-based video retrieval system requires first of all segment the video stream into separate shots. Video Shot Afterwards features are extracted for video shots representation. And finally, choose a similarity/distance metric and an algorithm that is efficient enough to retrieve query - related videos results. There are two main issues in this process; the first is how to determine the best way for video segmentation and key frame selection. The second is the features used for video representation. Various features can be extracted for this sake including either low or high level features. A key issue is how to bridge the gap between low and high level features. In this paper we presented approach for content based video retrieval based on Dominant color and texture of a video image. We also talk about video Representation, feature extraction from like texture, dominant color and color histogram from video frame. Keywords- Video retrieval, dominant color, Gray level co occurrence matrix. Feature extraction, Key frame extraction, Video representation, and Video segmentation. Image Retrieval, color Histogram

RTextTools: A Supervised Learning Package for Text Classification

Social scientists have long hand-labeled texts to create datasets useful for studying topics from congressional policymaking to media reporting. Many social scientists have begun to incorporate machine learning into their toolkits. RTextTools was designed to make machine learning accessible by providing a start-to-finish product in less than 10 steps. After installing RTextTools, the initial step is to generate a document term matrix. Second, a container object is created, which holds all the objects needed for further analysis. Third, users can use up to nine algorithms to train their data. Fourth, the data are classified. Fifth, the classification is summarized. Sixth, functions are available for performance evaluation. Seventh, ensemble agreement is conducted. Eighth, users can cross-validate their data. Finally, users write their data to a spreadsheet, allowing for further manual coding if required.

大規模データの俯瞰とターゲットデータの抽出に対する文書‐単語行列の特異値分解と特異値による重みづけの有効性

大規模データの俯瞰とターゲットデータの抽出に対する文書‐単語行列の特異値分解と特異値による重みづけの有効性

An Enhanced ILD Diagnosis Method using DWT

Interstitial Lung Disease (ILD) is a group of lung diseases affecting lung parenchyma. Since the lesions can be clearly identified in the CT scan of the lung, CT analysis is best among the ways for the identification of ILD by the pathologists. The automated detection of any disease from images uses the same fact that they are diagnosed by the medical professionals by exploiting the appearance features of the image of organ under consideration. Here we are applying the same to detect Lung Diseases from CT images. Segmenting the arteries and veins from the image is one of the major steps in analyzing the medical image. We need an edge enhanced lung image to do the same, so that the vascular tree can be clearly identified. In this implementation, discrete wavelet transform is applied for edge enhancement followed by the dynamic range compression. Wavelet edge enhancement and vessel enhancement filtering comprises the first stage of the algorithm. Vessel enhancement filtering uses the Eigen values of the Hessian of the image. The second stage of the algorithm corresponds to the feature extraction and classification. Feature extraction is done from the co- occurrence matrix of the resulting vessel segmented image. The co-occurrence features of the image forms the input feature vector for fuzzy SVM classifier. The performance of the proposed scheme is evaluated for accuracy.

Textural Feature Extraction and Classification of Mammogram Images using CCCM and PNN

This work presents and investigates the discriminatory capability of contourlet coefficient co- occurrence matrix features in the analysis of mammogram images and its classification. It has been revealed that contourlet transform has a remarkable potential for analysis of images representing smooth contours and fine geometrical structures, thus suitable for textural details. Initially the ROI (Region of Interest) is cropped from the original image and its contrast is enhanced using histogram equalization. The ROI is decomposed using contourlet transform and the co-occurrence matrices are generated for four different directions (θ=0°, 45°, 90° and 135°) and distance (d= 1 pixel). For each co-occurrence matrix a variety of second order statistical texture features are extracted and the dimensionality of the features is reduced using Sequential Floating Forward Selection (SFFS) algorithm. A PNN is used for the purpose of classification. For experimental evaluation, 200 images are taken from mini MIAS (Mammographic Image Analysis Society) database. Experimental results show that the proposed methodology is more efficient and maximum classification accuracy of 92.5% is achieved. The results prove that contourlet coefficient co-occurrence matrix texture features can be successfully applied for the classification of mammogram images. Keywords-Contourlet Transform, Mammogram, SFFS, PNN, ROI, MIAS

Content Based Medical Image Retrieval with Texture Content Using Gray Level Co-occurrence Matrix and K-Means Clustering Algorithms

Problem statement: Recently, there has been a huge progress in collection of varied image databases in the form of digital. Most of the users found it difficult to search and retrieve required images in large collections. In order to provide an effective and efficient search engine tool, the system has been implemented. In image retrieval system, there is no methodologies have been considered directly to retrieve the images from databases. Instead of that, various visual features that have been considered indirect to retrieve the images from databases. In this system, one of the visual features such as texture that has been considered indirectly into images to extract the feature of the image. That featured images only have been considered for the retrieval process in order to retrieve exact desired images from the databases. Approach: The aim of this study is to construct an efficient image retrieval tool namely, “Content Based Medical Image Retrieval with Texture Content using Gray Level Co-occurrence Matrix (GLCM) and k-Means Clustering algorithms”. This image retrieval tool is capable of retrieving images based on the texture feature of the image and it takes into account the Pre-processing, feature extraction, Classification and retrieval steps in order to construct an efficient retrieval tool. The main feature of this tool is used of GLCM of the extracting texture pattern of the image and k-means clustering algorithm for image classification in order to improve retrieval efficiency. The proposed image retrieval system consists of three stages i.e., segmentation, texture feature extraction and clustering process. In the segmentation process, preprocessing step to segment the image into blocks is carried out. A reduction in an image region to be processed is carried out in the texture feature extraction process and finally, the extracted image is clustered using the k-means algorithm. The proposed system is employed for domain specific based search engine for medical Images such as CT-Scan, MRI-Scan and X-Ray. Results: For retrieval efficiency calculation, conventional measures namely precision and recall were calculated using 1000 real time medical images (100 in each category) from the MATLAB Workspace database. For selected query images from the MATLAB-Image Processing tool Box-Workspace Database, the proposed tool was tested and the precision and recall results were presented. The result indicates that the tool gives better performance in terms of percentage for all the 1000 real time medical images from which the scalable performance of the system has been proved. Conclusion: This study proposed a model for the Content Based Medical Image Retrieval System by using texture feature in calculating the Gray Level Co Occurrence matrix (GLCM) from which various statistical measures were computed in order to increasing similarities between query image and database images for improving the retrieval performance along with the large scalability of the databases.

인자점수와 자기조직화지도를 이용한 희소한 문서데이터의 군집화

The retrieved documents have to be transformed into proper data structure for the clustering algorithms of statistics and machine learning. A popular data structure for document clustering is document-term matrix. This matrix has the occurred frequency value of a term in each document. There is a sparsity problem in this matrix because most frequencies of the matrix are 0 values. This problem affects the clustering performance. The sparseness of document-term matrix decreases the performance of clustering result. So, this research uses the factor score by factor analysis to solve the sparsity problem in document clustering. The document-term matrix is transformed to document-factor score matrix using factor scores in this paper. Also, the document-factor score matrix is used as input data for document clustering. To compare the clustering performances between document-term matrix and document-factor score matrix, this research applies two typed matrices to self organizing map (SOM) clustering.

Abstract 4544: On using textural information from longitudinal PET scans for predicting tumor response

Abstract Purpose: Early therapy response prediction, employing biomarkers such as [F-18]-fluorodeoxyglucose (FDG) followed with positron emission tomography (PET) for imaging the tumor metabolic activity, is a very actively researched topic. Traditionally, only the first order intensity based feature estimates extracted from the longitudinal scans are used for the response evaluations. Nonetheless, lesion heterogeneity computed from the higher order textures of the baseline scan prior to treatment has been recently found to be correlating well with the final outcome. The aim of this study is to establish the significance of tumor texture in follow up evaluations. Methods: 15 patients suffering from metastatic colorectal cancer and treated with cetuximab were selected for the study. Baseline FDG PET scans were acquired prior to treatment and the follow up scanned at 1 week after 1st dose of treatment was investigated for early response. One to five lesions were identified per patient. Intensity based features (IF) computed from the standardized uptake value (SUV) image include SUVmax (maximum activity), SUVmean (mean activity), SUVpeak (highest mean activity in 1mL sphere within tumor), tumor lesion volume (TLV) in mL, total lesion activity (TLA), TLA = SUVmean x TLV in SUVmL and total lesion energy (TLE), TLE = SUVmean^2 x TLV in SUV^2mL. Texture covariates (TF) estimated include 11 activity level run length matrix (ALRL) features, 11 activity level size zone matrix (ALSZ) features, 7 activity level co occurrence matrix (ALCM) features, 5 activity level difference matrix (ALDC) features and 19 activity level histogram (ALHS) features. Response indices (RI) were computed as the % change with respect to baseline after adding up covariates over all lesions per patient under study. The dynamic range of IF was investigated in baseline. A Cox proportional hazards regression model was fit with RIs and the time dependent receiver operating characteristics (ROC) was determined. Leave one-out cross validation (LOCV) was performed to investigate the fitting of the model. The concordance measure with the time to progression (TTP) was determined in each of the scenarios. Results: The baseline IF dynamic ranges are listed below representing (Median ± Standard Deviation, Minimum - Maximum) the respective covariates. SUVmax(6.33 ± 2.34, 2.40 - 12.28). SUVpeak(4.86 ± 2.01, 1.64 - 10.52). SUVmean(4.05 ± 1.52, 1.61 - 8.39). TLV(5.71 ± 155.91, 0.48 - 1182.42). TLA (29.58 ± 872.80, 2.11 - 6620.88). TLE (152.79 ± 5138.32, 3.55 - 38963.30). Cox model on LOCV with the IF only covariate set and the IF+TF set resulted in concordance of 0.6570 and 0.6764 resp. Conclusion: The results with the current data denote an added value in using textural information for therapy response evaluation. Moreover, therapy response could be predicted as early as 1 week after 1st dose of treatment from the feature estimates. Acknowledgements: This work is financially supported by KU Leuven's Concerted Research Action GOA/11/006, Research Foundation - Flanders (FWO), IWT Agency for Innovation by Science and Technology - Applied Biomedical Research (TBM) project 070717 and IBBT-KU Leuven Future Health Department. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research; 2012 Mar 31-Apr 4; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2012;72(8 Suppl):Abstract nr 4544. doi:1538-7445.AM2012-4544

Commentary: Diversity and uniformity of island floras

ISSN 1948-6596 news and update commentary Diversity and uniformity of island floras Studies of the ecological, evolutionary and bio- geographic patterns of island organisms have played an enormous role in concept shaping and theory forming in ecology, evolutionary science and biogeography. Numerous observations and descriptions of strange island beings, stunning evolutionary radiations and robust empirical for- mulae (e. g. on species–area relations) led to the insightful and influential syntheses of Mayr, Pre- ston, Carlquist, Williams, MacArthur and Wilson and others. Such works have been seminal for thousands of descriptive, explorative, hypothesis- driven and even experimental papers on island peculiarities, island evolution, interaction webs, and dispersion and distribution of species. Excel- lent reviews and textbooks on these subjects have been published with focus on each of these ap- proaches; today island ecology, evolution and bio- geography are a very vital and diverse part of ecology. The special issue “Comparative ecological research on oceanic islands”, edited by Christoph Kueffer and Jose-Maria Fernandez-Palacios in Per- spectives in Plant Ecology, Evolution and Sys- tematics (Vol. 12 iss. 2, April 2010), follows the tradition of tackling the subject from a focussed viewpoint. This time the focus is on island floras, and for once the scale is genuinely macroecologi- cal. All the papers involve as many islands or archi- pelagos as possible and attempt to make truly comparative analyses between them – to recog- nize patterns, and in some cases even to suggest explanations for the revealed pattern. The papers demonstrate that island ecology has come a long way from when it was a matter of finding strange plants or animals on remote islands (e.g. dodos), or amazing pedigrees in archipelago lineages (e.g. finches), or odd organismal interactions (e.g. ma- rine iguanas), or gross species numbers (“the holy S”). The editorial prelude by Kueffer and Fernandez-Palacios (2010), is a good introduction that tunes the ear to the harmonies and dishar- monies of island floras, as performed for us in the five subsequent articles. The main tune is that there are many ecological and biogeographical island and archipelago data and analyses avail- able, but few attempts at global syntheses. So, they argue, we should push “towards a common research agenda among biogeographers and ecologists in oceanic island research”. I applaud that. I certainly also appreciate the minor-tuned finale: we should do it before it is too late! The first two papers are truly island bio- geographic papers, focusing on species diversities. Chiarucci et al. (2010) aim to compare the floras of six well-known oceanic archipelagos. They try to take the very important step from species rich- ness (“the holy S”) to other measures of diversity. I find particularly interesting how they tackle beta diversity – so as to extract much more information from the species occurrence matrix than classic analyses of relations between island species num- ber (S) and area, or isolation, or altitude. Their approach is to make additive partitions between the alpha, beta and gamma diversities. This is a step forward but no leap: in their calculations they actually only apply total diversity for the entire archipelago (gamma), the mean S of the islands (alpha) and the difference between S max (the S of the richest island) and S i . In the species occur- rence matrix there is a wealth of information on beta diversity in the number of common species for each pair of islands; this information is not in- corporated in their approach. Whether step or leap, I find their paper stimulating and I am con- vinced that the last word on beta diversity in ar- chipelagos has not yet been uttered. Dominguez Lozano et al. (2010) compare the Canarian and Hawaiian archipelagos. Their comparisons concern both the entire floras and the floral composition of prominent habitat types. They consider taxonomic evenness, expressed as an index of the distribution the species within families and genera, and phylogenetic diversity (average taxonomic distinctness and the variance of the taxonomic distinctness). The authors con- vincingly show differences between the archipela- © 2010 the authors; journal compilation © 2010 The International Biogeography Society — frontiers of biogeography 2.3, 2010

Lexical Noise Analysis and Removal in Intelligent Search Engines

In the field of intelligent information retrieval (IR), latent semantic indexing (LSI) is a popular technique used to retrieve information related more in meaning than in lexical matching. This technique overcomes the problems associated with synonymy and polysemy (common causes of inaccuracy in matching algorithms). A core component in the process is the use of the singular value decomposition (SVD) which acts as a mathematical model for the lexical noise in the term document matrix (TDM). This paper investigates various aspects of LSI from the viewpoint of noise modeling and removal in image processing. A discussion about and an investigation into, mathematical modeling for lexical noise in the TDM is presented. The work addresses a definition for noise in text processing and seeks to determine the best structure of the TDM. In other words, the structure of the TDM that would facilitate efficient searching within the LSI.

Latent topic based multi-instance learning method for localized content-based image retrieval

Focusing on the problem of localized content-based image retrieval, based on probabilistic latent semantic analysis (PLSA) and transductive support vector machine (TSVM), a novel semi-supervised multi-instance learning (SSMIL) algorithm is proposed, where a bag corresponds to an image and an instance corresponds to the low-level visual features of a segmented region. In order to convert an MIL problem into a standard supervised learning problem, first, all the instances in training bags be clustered by K-Means method, and regards each cluster center as “visual-word” to build a visual vocabulary. Second, according to the distance between “visual-word” and instance, a fuzzy membership function is defined to establish a fuzzy term-document matrix, then use PLSA method to obtain bag’s (image’s) latent topic models, which can convert every bag to a single sample. Finally, in order to use the unlabeled images to improve retrieval accuracy, using semi-supervised TSVM to train classifiers. Experimental results on the COREL data sets show that the proposed method, named PLSA-SSMIL, is robust, and its performance is superior to other key existing MIL algorithms.

Improved Bags-of-Words Algorithm for Scene Recognition

This paper proposes a new bags-of-words (BoW)-based algorithm for scene/place recognition. Current scene recognition works that adopt BoW as the framework usually use a single codeword to represent the clusters obtained by k-means. Further, most of them often assign a hard value to a certain codeword to construct the BoW histogram. Using a single codeword to represent each cluster in fact is very preliminary since different clusters usually have different mean and covariance values. This causes using only mean value-based codeword will lose the covariance information and also makes the hard assignment to the codeword become biased. Considering this, this paper proposes an effective BoW-based technique to perform scene recognition. It first uses k-means algorithm to cluster the feature vectors into a certain number of clusters, in addition with an occurrence matrix. Gaussian mixed model (GMM) is then used to model the distribution of each cluster. Each GMM will be used as the new “codeword” of the codebook. Finally we propose to establish a new soft BoW histogram to represent each image through the soft assignment of the image features to each GMM. Support vector machine (SVM) is used to train these BoW histograms. Experimental results on the 15 categories dataset show that the proposed new BoW-based approach is very effective for scene/place recognition.