Searching for a specific person in family photos

• Main Author: Choo, Patrick Boon Choon.
• Other Authors: Teoh Eam Khwang
• Format: Final Year Project
• Language: English
• Published: 2012
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Control and instrumentation
• Online Access: http://hdl.handle.net/10356/49942
• Institution: Nanyang Technological University

This report provides a detailed documentation on the development of a SIFT approach face recognition system for identifying specific person in family photos. An in depth study and appreciation of the Scale Invariant Feature Transform (SIFT) and the Receiver Operating Curve (ROC) were involved in this project. Initially, fundamental steps in images processing were explored and implemented using Matlab in the development of the face recognition framework for digital images. Next, the “Public Figures Face” database which contains about 60 000 images for 200 public figures were acquired. 20 randomly selected public figures data sets were used to train and evaluate the face recognition system. Subsequently, five different SIFT models used for matching in the recognition stage were created for each of the selected public figures for comparison. Thereafter, ROC curve was used to determine the optimal threshold values that define the boundary of classification in face recognition so as to further enhance the system performance. Furthermore, a user friendly graphical user interface is created for the improvised face recognition framework. Finally, the SIFT approach was compared with the PCA approach implemented in a separate face recognition project. In this project, the SIFT approach face recognition framework performs at an average accuracy of 82%. In comparison, the PCA approach in a separate implementation performs at 66% accuracy.