DSpace Repository

Improving the Performance of Local Keypoint Descriptors for Image and Video Applications

Show simple item record

dc.contributor.advisor Dr. Nitin Afzulpurkar
dc.contributor.author Baber, Junaid
dc.contributor.other Dr. Matthew Dailey (Member) Prof. Sumanta Guha (Member)
dc.date.accessioned 2015-01-20T06:51:56Z
dc.date.available 2015-01-20T06:51:56Z
dc.date.issued 2013-12
dc.identifier.other AIT Diss no.CS-13-07 en_US
dc.identifier.uri http://www.cs.ait.ac.th/xmlui/handle/123456789/669
dc.description.abstract Test based search engines are getting famous, fast, and accurate despite the exponential growth in data. There are many search engines like Google, Bing and Yahoo which provides interactive and fast interface for searching; these searches are mainly based on text based key words and phrases. Now-a-days data and information is transformed in multimedia format such as images and videos. Multimedia contents are used for education, information and entertainment. These data particularly images are also increasing exponentially over Internet. People (users) are interested to query the search engine by some visual cues and retrieve the images and videos based on those cues, on the other hand, companies are interested to prevent the illegal distribution of forged and copyrighted images. There is much work reported for retrieval of images specially image copies, but still searching the query image or object is computationally very expensive, inaccurate and impractical like text based searches. The most famous criteria for image copy retrieval is to represent the images by set of local keypoint descriptors which should be robust, e cient, and discriminative. The problem to extract these descriptors is the main challenge. There are many descriptors but all descriptors lack either in e ciency, robustness or distinctiveness. Our framework increases the performance, particularly robustness and distinctiveness, of local keypoint descriptors for large scale image copy retrieval and video segmentation. We also propose binary quantization of gradient histograms which is very simple but e ective. This quantization can be applied for image copy retrieval for small scale and resource limited devices such as smart phones and tablets. This thesis describes the state-of-the art for image copy retrieval and video segmentation; we also present e ective techniques to improve the performance on challenging and benchmark datasets. en_US
dc.description.sponsorship University of Balochistan (UOB), Quetta, Pakistan- AIT Fellowship en_US
dc.publisher AIT en_US
dc.title Improving the Performance of Local Keypoint Descriptors for Image and Video Applications en_US
dc.type Dissertation en_US


Files in this item

This item appears in the following Collection(s)

Show simple item record

Search DSpace


Advanced Search

Browse

My Account