ABSTRACT

Feature detection in images is widely used in many applications, e.g feature matching, object recognition, and image retrieval. Several feature detectors algorithm have been proposed in the literature with a variety of functionalities. Th is work is focused on the performance evaluation of various features detectors with respect to different parameters. The feature detectors considered for the comparative analysis in this study include SURF, MSER, FAST, MinEigen and HARRIS. The parameters for performance evaluation are number of key features, execution time of the algorithm and feature matching accuracy. The dataset for this experiment is collected from previous research and discussed in Chapter No 3. It is based on Image objects such as sketch based plant objects, 3D photography objects, ear objects, office-home computer objects and office-home kettle object with different views are used for this experiment and analysis. It includes normal view; blur view, distorted view, scaling view and rotated 45 °, rotated 90°, rotated 180° and rotated 280 °. The simulations are performed with the comprehensive results which is shown and described with graphical representation and detailed discussion. For simulation purpose, MATLAB tool is used for study of the image feature detectors. This study shows that MinEigen, SURF and FAST functions extracts more no of key feature in normal view as compared to blur.
distorted, scaling and rotated views. They extracted low no of key features using home­ office-based object. Whereas FAST and MSER show high matching accuracy in normal view and other detector shows medium or less accuracy with respect to matching accuracy. The result of home-office kettle object MinEigen detector is extracted high no of key features in normal view and rotated 90°. FAST and MESER show good result than other detectors with respect to run time algorithm. SURF is computationally expensive as compared to rest of the detectors in normal view of ear object and sketched based plant object.