Songül Albayrak

Visualization and analysis of classifiers performance in multi-class medical data

The primary role of the thyroid gland is to help regulation of the bodys metabolism. The correct diagnosis of thyroid dysfunctions is very important and early diagnosis is the key factor in its successful treatment. In this article, we used four different kinds of classifiers, namely Bayesian, k-NN, k-Means and 2-D SOM to classify the thyroid gland data set. The robustness of classifiers with regard to sampling variations is examined using a cross validation method and the performance of classifiers in medical diagnostic is visualized by using cobweb representation. The cobweb representation is the original contribution of this work to visualize the classifiers performance when the data have more than two classes. This representation is a newly used method to visualize the classifiers performance in medical diagnosis.

Real time isolated turkish sign language recognition from video using hidden markov models with global features

This paper introduces a video based system that recognizes gestures of Turkish Sign Language (TSL). Hidden Markov Models (HMMs) have been applied to design a sign language recognizer because of the fact that HMMs seem ideal technology for gesture recognition due to its ability of handling dynamic motion. It is seen that sampling only four key-frames is enough to detect the gesture. Concentrating only on the global features of the generated signs, the system achieves a word accuracy of 95.7%.

Curvelet-Based Synthetic Aperture Radar Image Classification

Curvelet transform (CT) is a multiscale directional transform that enables the use of texture and spatial locality information. In synthetic aperture radar (SAR) imaging, CT is mostly used in speckle noise reduction. This letter utilizes CT for feature extraction in land use classification. Two types of curvelet-based feature extraction methods are implemented for SAR. The first one is defined and used in content-based image retrieval and is based on generalized Gaussian distribution parameter estimation for each curvelet subband. The second implementation is a genuine method that utilizes the use of curvelet subband histograms, namely, histogram of curvelets (HoC). Using the proposed curvelet-based feature extraction method (HoC) on SAR data, better classification accuracies up to 99.56% are achieved compared to original data and H/A/α decomposition features. Compared to speckle-noise-reduced data classification results, it can be said that curvelet-based feature extraction is also robust against speckle noise.

Turkish fingerspelling recognition system using Generalized Hough Transform, interest regions, and local descriptors

This paper presents a computer vision system that can recognize Turkish fingerspelling sign hand postures by a method based on the Generalized Hough Transform, interest regions, and local descriptors. A novel method for calculating the reference point for the Generalized Hough Transform, and a simpler but more effective Hough voting strategy are proposed. The stages of implementing a Generalized Hough Transform are examined in detail, and the issues that affect the method success are discussed. The system is tested on a data set with 29 classes of non-rigid hand postures signed by three different signers on non-uniform backgrounds. It attains a 0.93 success rate.

Turkish fingerspelling recognition system using axis of least inertia based fast alignment

Fingerspelling is used in sign language to spell out names of people and places for which there is no sign or for which the sign is not known. In this work we describe a Turkish fingerspelling recognition system that recognizes all 29 letters of the Turkish alphabet. A single representative frame is extracted from the sign video, since that frame is enough for recognition purposes of the letters mentioned. Processing a single frame, instead of the whole video, increases speed considerably. The skin regions in the representative frame are extracted by color segmentation in YCrCb space before clearing noise regions by morphological opening. A novel fast alignment method that uses the angle of orientation between the axis of least inertia and y axis is applied to hand regions. This method compensates small orientation differences but increases big ones. This is desirable when differentiating the fingerspelling signs, some of which are close in shape but different in orientation. Also the use of minimum bounding square is advised, which helps in resizing without breaking the alignment. Binary values of this minimum bounding square are directly used as feature values, and that allowed experimenting with different classification schemes. Features like mean radial distance and circularity are also used for increasing success rate. Classifiers like kNN, SVM, Naive Bayes, and RBF Network are experimented with, and 1NN and SVM are found to be the best two of them. The video database was created by 3 different signers, a set of 290 training videos, and a separate set of 174 testing videos are used in experiments. The best classifiers 1NN and SVM achieved a success rate of 99.43% and 98.83% respectively.

Turkish Sign Language recognition using spatio-temporal features on Kinect RGB video sequences and depth maps

This paper presents a Turkish Sign Language recognition system that uses spatio-temporal features on Kinect sensor RGB video sequences and depth maps. Proposed system uses cumulative motion images which based on motion differences and represent the temporal characteristics of dynamic signs in motion sequences. Cumulative motion images represent the whole motions of signers. 2-D Discrete Cosine Transform (DCT) is applied to cumulative sign images in order to obtain spatial features of signs and transformed images that represent the energy density of signs are obtained. Two transform images are obtained by applying referred methods to both of RGB video sequences and depth maps seperately. Feature vectors of dynamic signs are produced by combining a certain amount of DCT coefficients that contain higher energy via zig-zag scanning on transform images. K-Nearist Neighbor classifier with Manhattan distance used for recognition process. System performance is evaluated on a sign database that contains 1002 signs belongs to 111 words in three different categories of Turkish Sign Language (TID). Proposed sign language recognition system has a recognition rate about %90.

Comparison of feature selection algorithms for medical data

Data mining application areas widen day by day. Among those areas medical area has been receiving quite a big attention. However, working with very large data sets with many attributes is hard. Experts in this field use heavily advanced statistical analysis. The use of data mining techniques is fairly new. This paper compares three feature selection algorithms on medical data sets and comments on the importance of discretization of attributes.

Unsupervised clustering methods for medical data: an application to thyroid gland data

The purpose of this paper is to examine the unsupervised clustering methods on medical data. Neural networks and statistical methods can be used to develop an accurate automatic diagnostic system. Self-Organizing Feature map as a Neural Network model and K-means as a statistical model are tested to predict a well defined class. To test the diagnostic system, thyroid gland data is used for the application. As a result of clustering algorithms, patients are classified normal, hyperthyroid function and hypothyroid function.

Segmentation of Bone with Region Based Active Contour Model in PD Weighted MR Images of Shoulder.

Proton density (PD) weighted MR images present inhomogeneity problem, low signal to noise ratio (SNR) and cannot define bone borders clearly. Segmentation of PD weighted images is hampered with these properties of PD weighted images which even limit the visual inspection. The purpose of this study is to determine the effectiveness of segmentation of humeral head from axial PD MR images with active contour without edge (ACWE) model. We included 219 images from our original data set. We extended the use of speckle reducing anisotropic diffusion (SRAD) in PD MR images by estimation of standard deviation of noise (SDN) from ROI. To overcome the problem of initialization of the initial contour of these region based methods, the location of the initial contour was automatically determined with use of circular Hough transform. For comparison, signed pressure force (SPF), fuzzy C-means, and Gaussian mixture models were applied and segmentation results of all four methods were also compared with the manual segmentation results of an expert. Experimental results on our own database show promising results. This is the first study in the literature to segment normal and pathological humeral heads from PD weighted MR images.

User based and item based collaborative filtering with temporal dynamics

Collaborative Filtering or recommender systems use a database for new users and new items about theirs preferences. It is very important to make private suggestions to users, keep their interest alive with admirable suggestions. Collaborative Filtering (CF) is a commonly used system to meet this end. However, despite the fact that CF systems are widely used, traditional CF techniques are unable to track the preferences of users over a period of time. For this reason, “temporal dynamics” has become an important notion in recommendation systems. In this study a new method has been employed to provide customized suggestions to users whose tastes may have changed over time. The proposed system is different from the traditional user-based CF technique and item-based CF technique in that it examines the dates users ranked products and uses this data to help determine user preferences. The evaluation process has been performed on Netflix data in order to measure the success of the system and compare the results with traditional user-based CF technique and item-based C technique. The results are encouraging and the quality of the predictions were significantly improved.