Scale-Invariant Feature Transform based Single-Sample Face and Ear Recognition

Recognising a person with just one face or ear image is a challenging sub-problem in the field of biometric recognition. The difficulty arises as parameter estimation becomes challenging from a single image. This article proposes a methodology for recognition from a single face or ear image using scale-invariant feature transform (SIFT) for feature extraction. The proposed method is tested on five public databases, namely ORL, Yale, and AR (for facial images) and AMI and IITD (for ear images). Extensive image processing attacks like scaling (0.1 to 3.0), rotation (30 degrees to 330 degrees), and blurring (0.5 to 4.0) are applied to the test image to evaluate the performance of the method in real-time scenarios. The proposed method achieves excellent classification accuracy and shows robustness towards scaling, rotation, and blurring. The classification of the test image is done using the K-nearest neighbour (KNN) by measuring the Euclidean distance. The classification accuracy results are in the range of 88.86%-96.78% and 73.81%-90.50% for face and ear databases, respectively. The classification accuracy is in the range of 82.38%-96.78% and 81.30%-91.56% against scaling and blurring. Further, the proposed method shows acceptable classification accuracy in the range of 73.81%-95.56% on rotation as well. This proves that feature extraction methods like SIFT can be effective even in difficult scenarios where scarcity or only a few samples are available.