A comparative study on chrominance based methods in dorsal hand recognition: Single image case

Abstract

Dorsal hand recognition is a crucial topic in biometrics and human-machine interaction; however most of the identification systems identify and segment the hands from the images consisting of high contrast backgrounds. In other words, capturing and analyzing images of hands on a white or black or any colored background is way too easy to achieve high accuracy. On the contrary, in continuous authentication or in interactive human-machine systems, it can be not possible nor feasible to process high contrast images, like hands on computer keyboards which is not as simple as single color backgrounds even the feature to be extracted is solely the hand color. Therefore we deal with processing of the images consisting of hands on computer keyboards to compare various luminance and chrominance methods by YCbCr color space extraction and to find ways to achieve higher accuracy without any succeeding erosion, dilation or filtering. The methods focused on chromatic intervals could be summarized as: fixed intervals, covariance intervals and fuzzy 2-means. Our main contribution briefly is a necessary accuracy comparison and validation of the common methods on the single images. The highest accuracy is found as 96% by fuzzy 2-means applied to chrominance layers of the image.