Recognition using SIFT and its Variants on Improved Segmented Iris

Iris is one of the most reliable biometric traits due to its stability and randomness. Iris is transformed to polar coordinates by the conventional recognition systems. They perform well for the cooperative databases, but the performance deteriorates for the non-cooperative irises. In addition to this, aliasing effect is introduced as a result of transforming iris to polar domain. In this thesis, these issues are addressed by considering annular iris free from noise due to eyelids. This thesis presents several SIFT based methods for extracting distinctive invariant features from iris that can be used to perform reliable matching between different views of an object or scene. After localization of the iris, Scale Invariant Feature Transform (SIFT) is used to extract the local features. The SIFT descriptor is a widely used method for matching image features. But SIFT is found out to be computationally very complex. So we use another keypoint descriptor, Speeded up Robust Features (SURF), which is found to be computationally more efficient and produces better results than the SIFT. Both SIFT and SURF has the problem of false pairing. This has been overcome by using Fourier transform with SIFT (called F-SIFT) to obtain the keypoint descriptor and Phase-Only Correlation for feature matching. F-SIFT was found to have better accuracy than both SIFT and SURF as the problem of false pairing is significantly reduced. We also propose a new method called S-SIFT where we used S Transform with SIFT to obtain the keypoint descriptor for the image and Phase-Only Correlation for the feature matching. In the thesis we provide a comparative analysis of these four methods (SIFT, SURF, F-SIFT, S-SIFT) for feature extraction in iris.