Caroline Cao (Committee Member), Arthur Goshtasby (Advisor), Jack Jean (Committee Member), Thomas Wischgoll (Committee Member)
Doctor of Philosophy (PhD)
Local image features are used in many computer vision applications. Many point detectors and descriptors have been proposed in recent years; however, creation of effective descriptors is still a topic of research. The Scale Invariant Feature Transform (SIFT) developed by David Lowe is widely used in image matching and image retrieval. SIFT detects interest points in an image based on Scale-Space analysis, which is invariant to change in image scale. A SIFT descriptor contains gradient information about an image patch centered at a point of interest. SIFT is found to provide a high matching rate, is robust to image transformations; however, it is found to be slow in image matching/retrieval. Autoencoder is a method for representation learning and is used in this project to construct a low-dimensional representation of a high-dimensional data while preserving the structure and geometry of the data. In many computer vision tasks, the high dimensionality of input data means a high computational cost. The main motivation in this project is to improve the speed and the distinctness of SIFT descriptors. To achieve this, a new descriptor is proposed that is based on Autoencoder. Our newly generated descriptors can reduce the size and complexity of SIFT descriptors, reducing the time required in image matching and image retrieval.
Department or Program
Department of Computer Science and Engineering
Year Degree Awarded
Copyright 2016, all rights reserved. My ETD will be available under the "Fair Use" terms of copyright law.