Texture segmentation by clustering of Gabor feature vectors

Segmentation of gray level images into regions of uniform texture is investigated. An unsupervised approach through the use of Kohonen's self organizing map (SOM) and a multilayer version of it, the hierarchical self organizing map (HSOM), is employed to find the regions in an image composed of textures from different classes. For testing, gray level artificial textured images modeled as Markov random fields are used as the input. No parameter estimation is done. The size and the topology of SOM and HSOM are independent from the size of the input image. The segmentation results are very promising.

Target recognition by texture segmentation algorithm

In this work, a Recurrent Neural Network (RNN) is trained using cepstral features and a set of difference cepstral feature (DCF) vectors on a frame to frame basis. The DCF vector is formulated to capture the temporal patterns of fricative sounds or phonemes of Assamese language. A hybrid algorithm is developed to recognize these fricative phonemes from certain words containing them. To preserve the temporal information of the speech segment, we here consider a frame-based hybrid approach to recognize fricatives from Assamese speech. A hybrid feature set is developed where simple frame-based feature is combined with differential frame-based feature. Investigation of feature extraction techniques like Linear Predictive Cepstral Coefficient (LPCC) and Mel-Frequency Cepstral Coefficient (MFCC) have been carried out and their performances have been evaluated in comparison to that obtained from the DCF set for Assamese fricative phoneme recognition. Here, speech segment is divided into 20 millisecond frames with overlap of 10 millisecond to extract features. Also, difference of a current frame with its preceding and succeeding frame is considered for forming a more accurate dynamic approach for fricative recognition. The differential processing enables to reduce correlation and retain only the most relevant portion of the input. After obtaining the feature vectors, Self Organizing Map (SOM) has been used to categorize the related features into different classes and remove repeating data. Thus the features obtained from the phoneme signal has been reduced into different sized cluster centres provided by SOM. The reduced feature vector is next applied to the RNN based hybrid classifier for learning the pattern and recognizing any unknown fricative segment.

<b><sc>Abstract.</sc></b> The area irrigated by furrow irrigation in the U.S. has been steadily decreasing but still represents about 20% of the total irrigated area in the U.S. Furrow irrigation sediment loss is a major water quality issue in the western U.S. and a method for estimating sediment loss is needed to quantify the environmental impacts and estimate effectiveness and economic value of conservation practices. The objective of the study was to investigate the use of the unsupervised machine learning technique Kohonen self-organizing maps (KSOM) to predict furrow sediment loss. Historical published and unpublished data sets containing measurements of furrow irrigation sediment loss in the western U.S. were assembled into a furrow sediment loss data set comprising over 2000 measurements. Despite the immunity of KSOMs to measurement variability, the inherent variability in measured furrow sediment loss limited the ability of a KSOM model to reliability predict furrow sediment loss. Furrow sediment loss was under predicted by 44% on average with a linear regression coefficient of determination of 0.6. The KSOM model was placing little weight on measured sediment loss in the input data set, indicating that it was clustering the data based on input parameters defining the hydraulic and soil conditions. This outcome was used to develop a transfer learning approach for predicting furrow sediment loss. The transfer learning approach used a KSOM to cluster data records of similar hydraulic and soil conditions in the data set. Mean measured sediment loss and furrow flow rate of each cluster was determined based on data set vectors assigned to a cluster by the KSOM. Furrow sediment loss prediction was obtained by applying an input vector to the KSOM to identify the cluster the input vector most closely matches. Then the mean measured sediment loss of the identified cluster was adjusted for any difference between the input vector furrow flow rate and cluster mean furrow flow rate to obtain a prediction of furrow sediment loss. Predicted furrow sediment loss was 16% less than measured sediment loss on average with a coefficient of determination of 0.82. When the data set was randomly split into model development (90%) and validation (10%) data sets the prediction results were similar.

A fuzzy self-organizing map (SOM) network is proposed in this paper for visual motor control of a 7 degrees of freedom (DOF) robot manipulator. The inverse kinematic map from the image plane to joint angle space of a redundant manipulator is highly nonlinear and ill-posed in the sense that a typical end-effector position is associated with several joint angle vectors. In the proposed approach, the robot workspace in image plane is discretized into a number of fuzzy regions whose center locations and fuzzy membership values are determined using a Fuzzy C-Mean (FCM) clustering algorithm. SOM network then learns the inverse kinematics by on-line by associating a local linear map for each cluster. A novel learning algorithm has been proposed to make the robot manipulator to reach a target position. Any arbitrary level of accuracy can be achieved with a number of fine movements of the manipulator tip. These fine movements depend on the error between the target position and the current manipulator position. In particular, the fuzzy model is found to be better as compared to Kohonen self-organizing map (KSOM) based learning scheme proposed for visual motor control. Like existing KSOM learning schemes, the proposed scheme leads to a unique inverse kinematic solution even for a redundant manipulator. The proposed algorithms have been successfully implemented in real-time on a 7 DOF PowerCube robot manipulator, and results are found to concur with the theoretical findings.

In the past few years, image retrieval has been one of the hot spots in computer vision field. Among many image retrieval techniques, Bag-of, Word (BoW) model is one of the effective and efficient methods that can search images with visual vocabularies and it is insensitive to massive data and various geometric attacks. But the classical BoW algorithm used some descriptors as its visual words, such as SIFT, which are also used to build visual vocabulary. The main problem with traditional BoW algorithm is that the visual vocabulary could not reflect the spatial information of visual words. And most BoW algorithms utilize one single feature as their feature vector, some other important features are ignored. All these factors have influenced the accuracy of final result in traditional BoW model. In order to solve the problem, we propose a novel image retrieval method with BoW and Gabor feature. The paper first proposes a new saliency map extraction method, and then the saliency score is used to build visual vocabulary. At last, Gabor feature is combined with visual vocabulary together to compute similarity. In order to test the effectiveness of the algorithm, we evaluate our method on the SIMPLICITY dataset and Stanford dataset. Experiments results demonstrate the effectiveness and accuracy of the proposed method.

QuestionCan a Kohonen's self‐organizing map, which is robust to non‐linear relationships between variables and their non‐normal distributions, be effective in patterning the data on plant communities investigated with the classical Braun‐Blanquet phytosociological method based on the ordinal scale? Does the application of the self‐organizing map make it possible to obtain new information from the analysed plant communities when compared to the Braun–Blanquet method alone?LocationThe Babsk nature reserve, Central Poland.MethodsThe analysed data were derived from two separate series of phytosociological studies on plant communities dating from periods 31 yr apart (1960 and 1991). The data consisted of 24 quantitative sampling lists of plant species (=phytosociological releves). The releves were analysed by the application of a Kohonen's self‐organizing map and the indicator value (IndVal).ResultsThe transformations from Querco roboris–Pinetum to Tilio cordatae–Carpinetum betuli in the vegetation in the Babsk nature reserve over 31 yr were determined precisely. Valuable new data were also obtained on: (1) significant associations of individual plant species with the previous and recent phytocoenoses, i.e. diagnostic groups of plant species for each phytocoenosis; (2) abiotic conditions (determined retrospectively in the two study periods on the basis of ecological indicators for vascular plants); and (3) the concordance of phytocoenoses with the biotope.ConclusionsThe Kohonen's self‐organizing map method and the IndVal made it possible to efficiently identify abiotic and biotic patterns for plant communities on the basis of the data expressed in the conventional Braun‐Blanquet scale.

Traditional texture region location methods with Gabor features are often limited in the selection of Gabor filters and fail to deal with the target which contains both texture and non-texture parts. Thus, to solve this problem, a two-step new model was proposed. In the first step, the original features extracted by Gabor filters are applied to training a self-organizing map (SOM) neural network and a novel merging scheme is presented to achieve the clustering. A back propagation (BP) network is used as a classifier to locate the target region approximately. In the second step, Chan-Vese active contour model is applied to detecting the boundary of the target region accurately and morphological processing is used to create a connected domain whose convex hull can cover the target region. In the experiments, the proposed method is demonstrated accurate and robust in localizing target on texture database and practical barcode location system as well.

As the coverage area of the sub-block is too large, the texture feature clustering based on sub-block often produces the mosaic phenomenon of inaccurate boundary. In this paper, the texture clustering algorithm based on pixel extracts the texture feature vector of its central pixel point from sub-block. The image texture feature is standardized, then the normalized feature vector is clustered and the clustering result is used to realize the segmentation of complex image. Compared with the traditional segmentation methods such as gradient method, threshold method and so on, the segmentation boundary is accurate, and the phenomenon of under-segmentation and over-segmentation also reduce significantly.

Video compression is an essential task in video storage and transmission. In this paper, we worked for compressing the video sequence using Self-organizing map (SOM) and then stored the output feature vector of SOM using Hopfield Neural Network. The SOM algorithm is based on unsupervised, competitive learning. It extracts the useful features from video frames by feature extraction method. In our approach, the video sequence is first divided into video frames. The frames are processed by passing to the input of Self-organizing map (SOM) that outputs the informative features by removing some of the redundant information contained in video frames. Self-organizing map is an efficient method for reducing the size of images in the neural network field. The feature vector obtained is further used for storing the patterns using Hopfield network. The feature vector is applied to the input of the Hopfield neural network and stored in the network for encoding the video frames. Thus the compressed video frames are encoded and stored in memory as a codeword. The codeword is used for reconstruction of the video frames by using the same Hopfield network. The codeword is applied to the Self-organizing map network and the video frames are reconstructed from those codeword. Simulation results show that high compression rate is achieved while maintaining the good reconstruction quality of images.

This paper is concerned with identifying and estimating the parameters of the different texture regions that comprise a textured image. A textured region here is modelled by a Markov Random Field (MRF). The MRF is parametrized by a parameter vector &alpha; , ana has a noncausal structure. We assume no a prior knowledge about the different texture regions, their associated texture parameters, or the available number of textured regions. The image is partitioned into disjoint square windows and a maximum likelihood estimate (MLE) (or a sufficient statistis) &alpha;* for &alpha; (for a fixed order model) is obtained in each window. The components of &alpha;* are viewed as features, and a as a feature vector. The windows are grouped in different texture regions based on feature selection and clustering analysis of the &alpha;* vectors in the different windows. To simplify the clustering process, the dimensionality of the feature vector is reduced via a Karhunen-Loeve decomposition of the between-to-within scatter matrix of the &alpha;* vectors. Each &alpha;* is projected onto the dominant mode (eigenvector) of the scatter matrix. The projected data is used in the clustering process. The clustering is achieved by minimizing a within group variance criterion which has been weighted by a factor that explicitly depends on the number of groups. To reduce the computational cost associated with this method, it is accompanied by a &quot;valley method&quot;. Finally, by exploiting the asymptotic normality of the MLE, we compute the tglobal MLE &alpha;* for each textured region by properly combining the locally estimated MLE &alpha;* in the various windows that comprise the region. The global MLE &alpha;* for a region is notning but an appropriately weighted linear combination of the local MLE set {^&alpha;k^*}.

Describes a series of experiments on using Kohonen self-organizing maps and hybrid systems for continuous speech recognition. Experiments with different nonlinear transformations on the signal before using a neural network has been done and results compared. The hybrid system developed by the authors combines self-organizing feature maps with dynamic time warping. The experiments suggest that the combination has better performance than either of the two methods applied individually. >

To extract fuzzy rules from databases or data files, we first select a set of attributes to associate and restrict all records (rows) to these. We next embed these feature vectors of small dimension into a high dimensional feature space by a Gaussian kernel mapping that yields a symmetric fuzzy membership matrix. The entry value at row i and column] is a fuzzy truth value that feature vectors i and j are in the same cluster. We look for clusters where features A and B associate in some way, e.g., (A is HIGH) and (B is LOW), so if the support and confidence are high enough, we accept that rule. The cluster centers become centers of fuzzy set membership functions to use in fuzzy modus ponens with the fuzzy rules . We apply our novel algorithm to analyze two difficult well known datasets.

Stock market prediction is a very noisy problem and the use of any additional information to increase accuracy is necessary. In this paper, for the stock daily return prediction problem, the set of features is expanded to include indicators not only for the stock to be predicted itself but also a set of other stocks and currencies. Afterwards, different feature selection and classification methods are utilized for prediction. The daily close returns of the 3 most traded stocks (GARAN, THYAO, and ISCTR) in Borsa Istanbul (BIST) are predicted using indicators computed on those stocks, indicators for all the other stocks listed in the BIST100 index, and indicators on the dollar-gold prices. Twenty-five different indicators on daily stock prices are computed to form feature vectors for each trading day. These feature vectors are assigned class labels according to the daily close returns. Expanding the feature space with BIST100 stocks features results in a high dimensional feature space, with possibly noisy or irrelevant features. Therefore, feature selection methods are utilized to select the most informative features. In order to determine relevance scores of features, fast filter-based methods, gain ratio and relief, are used. Experiments are performed based on individual stock features, dollar-gold features (DG), BIST100 stock features (BIST100), and a combination of BIST100 and DG with and without feature selection. Using the gain ratio feature selection with a gradient boosting machine (GBM), the movements of GARAN stock were predicted with an accuracy of 0.599 and an F-measure of 0.614. For THYAO, the relief feature selection with the GBM gave an accuracy of 0.558, and for ISCTR, the gain ratio feature selection with logistic regression achieved an accuracy of 0.581. It was found that using BIST100 stock features boosts classification results for all stocks in terms of accuracy.

There is a trend in recent machine learning community to construct a nonlinear version of a linear algorithm using the 'kernel method', e.g. Support Vector Machines (SVMs), kernel principal component analysis, kernel fisher discriminant analysis and the recent kernel clustering algorithms. In unsupervised clustering algorithms using kernel method, typically, a nonlinear mapping is used first to map the data into a potentially much higher feature space, where clustering is then performed. A drawback of these kernel clustering algorithms is thatthe clustering prototypes lie in high dimensional feature space and hence lack clear and intuitive descriptions unless using additional projection approximation from the feature to the data space as done in the existing literatures. In this paper, a novel clustering algorithm using the 'kernel method' based on the classical fuzzy clustering algorithm (FCM) is proposed and called as kernel fuzzy c-means algorithm (KFCM). KFCM adopts a new kernel-induced metric in the data space to replace the original Euclidean norm metric in FCM and the clustered prototypes still lie in the data space so that the clustering results can be reformulated and interpreted in the original space. Our analysis shows that KFCM is robust to noise and outliers and also tolerates unequal sized clusters. And finally this property is utilized to cluster incomplete data. Experiments on two artificial and one real datasets show that KFCM hasbetter clustering performance and more robust than several modifications of FCM for incomplete data clustering.