mussacharles / ear-recognition-system-using-chainlets-for-personal-identification Goto Github PK

This project forked from andudu/ear-recognition-system-using-chainlets-for-personal-identification

The stable biometric nature of the ear has led to an interest in ear recognition. The advantages of fixed shape and complex but stable appearance of makes it a potentially reliable biometric. We proposed the use of CHAINLETS using Deep Contour-based edge detection as a novel ear descriptor. CHAINLETS supports patches of Histogram of Chain Code to describe an ear and support accurate biometric ear matching. These descriptors significantly outperform existing feature sets for human ear recognition. Our study of the existing edge boundary and Chain Code based descriptors played a significant part in determining the direction of ear edges, especially when applied on patches of an ear. We performed an extensive evaluation using highquality local contrast normalization in overlapping descriptor blocks and the similarity distances for matching on new Unconstrained Ear Recognition Challenge dataset from the University of Ljubljana, Slovenia. The experimental results show that the novel CHAINLETS approach achieves higher recognition performance than the state-of-the-art algorithms.