Investigation of sift for gender classification

As per the recent study, the study of face images tends to dominate the research field on Gender classification. However, using face images as the primary gender classifier is possible only after considering the assumptions such as frontal face and favorable illumination conditions. Further...

Full description

Saved in:
Bibliographic Details
Main Author: Harihara Subramanian Karthikk.
Other Authors: Sung, Eric
Format: Theses and Dissertations
Language:English
Published: 2013
Subjects:
Online Access:http://hdl.handle.net/10356/54701
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:As per the recent study, the study of face images tends to dominate the research field on Gender classification. However, using face images as the primary gender classifier is possible only after considering the assumptions such as frontal face and favorable illumination conditions. Further, more real-time problems such as pose, background clutter adds up to the complexities towards the approach. Also the results get affected when there is a change in scale and rotation. Through this dissertation work, it is sought to tackle the above mentioned problems by eliminating them. These complexities are addressed by SIFT keypoint vector algorithm which narrows down the approach for classifying gender. To support this claim, an investigation on a new approach to classify facial images based on gender was done. Scale Invariant Feature Transform (SIFT) is one of the most popular local image descriptors in use. The SIFT vectors extracted from a sample database is examined for any distinctive characteristics. Stanford University Medical Student (SUMS) frontal facial image database was used for testing the performance. Both the location and 128-bit descriptor vector of the SIFT keypoint are individually analyzed for the investigation. All the SIFT keypoints generated from a small database of male and female face images are analyzed to find some distinctive keypoint vectors which can differentiate between the two classes face images. It has been found that the SIFT keypoint vectors are not robust enough for accurate classification. Hence, dense SIFT concept is used to improve the efficiency rate.