Erkan Bostanci

An Evaluation of Classification Algorithms Using Mc Nemar’s Test

Five classification algorithms namely J48, Naive Bayes, Multilayer Perceptron, IBK and Bayes Net are evaluated using Mc Nemar’s test over datasets including both nominal and numeric attributes. It was found that Multilayer Perceptron performed better than the two other classification methods for both nominal and numerical datasets. Furthermore, it was observed that the results of our evaluation concur with Kappa statistic and Root Mean Squared Error, two well-known metrics used for evaluating machine learning algorithms.

Augmented reality applications for cultural heritage using Kinect

This paper explores the use of data from the Kinect sensor for performing augmented reality, with emphasis on cultural heritage applications. It is shown that the combination of depth and image correspondences from the Kinect can yield a reliable estimate of the location and pose of the camera, though noise from the depth sensor introduces an unpleasant jittering of the rendered view. Kalman filtering of the camera position was found to yield a much more stable view. Results show that the system is accurate enough for in situ augmented reality applications. Skeleton tracking using Kinect data allows the appearance of participants to be augmented, and together these facilitate the development of cultural heritage applications.

A genetic algorithm solution to the collaborative filtering problem

A genetic algorithm based solution for the collaborative filtering was proposed.This model was tested on weights computed with different similarity metrics.The performance of different metrics after evolutionary approach was compared. Development of approaches for reducing the prediction error has been an active research field in collaborative filtering recommender systems since the accuracy of the prediction plays a crucial role in user purchase preferences. Unlike the conventional collaborative filtering methods which directly use the computed user-to-user similarity values, this paper presents a genetic algorithm approach for refining them before using in the prediction process. The approach was found to yield promising results according to the statistical analysis performed on a variety numbers of neighbours for various similarity metrics including Pearsons Correlation, Extended Jaccard Coefficient and Vector Cosine Similarity along with a metric that assigns random weights to be used as a benchmark. Results show that the evolutionary approach has significantly reduced the prediction error using the evolved weights and Vector Cosine Similarity has shown the best performance.

Spatial Statistics of Image Features for Performance Comparison

When matching images for applications such as mosaicking and homography estimation, the distribution of features across the overlap region affects the accuracy of the result. This paper uses the spatial statistics of these features, measured by Ripleys K-function, to assess whether feature matches are clustered together or spread around the overlap region. A comparison of the performances of a dozen state-of-the-art feature detectors is then carried out using analysis of variance and a large image database. Results show that SFOP introduces significantly less aggregation than the other detectors tested. When the detectors are rank-ordered by this performance measure, the order is broadly similar to those obtained by other means, suggesting that the ordering reflects genuine performance differences. Experiments on stitching images into mosaics confirm that better coverage values yield better quality outputs.

A Navigation System for the Visually Impaired: A Fusion of Vision and Depth Sensor

For a number of years, scientists have been trying to develop aids that can make visually impaired people more independent and aware of their surroundings. Computer-based automatic navigation tools are one example of this, motivated by the increasing miniaturization of electronics and the improvement in processing power and sensing capabilities. This paper presents a complete navigation system based on low cost and physically unobtrusive sensors such as a camera and an infrared sensor. The system is based around corners and depth values from Kinects infrared sensor. Obstacles are found in images from a camera using corner detection, while input from the depth sensor provides the corresponding distance. The combination is both efficient and robust. The system not only identifies hurdles but also suggests a safe path (if available) to the left or right side and tells the user to stop, move left, or move right. The system has been tested in real time by both blindfolded and blind people at different indoor and outdoor locations, demonstrating that it operates adequately.

Is Hamming distance only way for matching binary image feature descriptors

Brute force matching of binary image feature descriptors is conventionally performed using the Hamming distance. The use of alternative metrics is assessed in order to see whether they can produce feature correspondences that yield more accurate homography matrices. Two statistical tests, namely analysis of variance (ANOVA) and McNemars test were employed for the evaluation. Results show that Jackard-Needham and Dice metrics can display better performance for some descriptors. Yet, these performance differences were not found to be statistically significant.

User Tracking Methods for Augmented Reality

Augmented reality has been an active area ofresearch for the last two decades or so. This paper presents acomprehensive review of the recent literature on trackingmethods used in Augmented Reality applications, both forindoor and outdoor environments. After critical discussion ofthe methods used for tracking, the paper identifies limitations ofthe state-of-the-art techniques and suggests potential futuredirections to overcome the bottlenecks.

Extracting planar features from Kinect sensor

An algorithm for finding planar features from a 3D point cloud by Kinects depth sensor is described in this paper. The algorithm uses the explicit definition of a plane which allows storing only four parameters per plane rather than storing thousands of points. Extraction of multiple planes from the same set of points is prevented using a rejection mechanism. Parallelism is used for an average speed-up of 2.3:1. Details of the algorithm and results are given along with a discussion of how the calibration of the sensor affects the projections.

Vision-based user tracking for outdoor augmented reality

This paper examines the use of vision-based localization techniques for indoor environments in outdoor environments. A new method is presented for robust data association and finding camera trajectory; based on these, a simple augmented reality game is implemented.

Evaluation Method, Dataset Size or Dataset Content: How to Evaluate Algorithms for Image Matching?

Most vision papers have to include some evaluation work in order to demonstrate that the algorithm proposed is an improvement on existing ones. Generally, these evaluation results are presented in tabular or graphical forms. Neither of these is ideal because there is no indication as to whether any performance differences are statistically significant. Moreover, the size and nature of the dataset used for evaluation will obviously have a bearing on the results, and neither of these factors are usually discussed. This paper evaluates the effectiveness of commonly used performance characterization metrics for image feature detection and description for matching problems and explores the use of statistical tests such as McNemar’s test and ANOVA as better alternatives.