Exploring Circular Hough Transforms for Detecting Hand Feature Points in Noisy Images from Ghost-Circle Patterns

Abstract

Several applications involve the automatic analysis of hand images such as biometry, digit-ratio measurements, and gesture recognition. A key problem common to these applications is the separation of hands from the background. Color based approaches struggle to detect the boundaries of the hand and the background if these have similar colors. This paper thus describes work-in-progress with a spatial approach for finger feature point detection based on the circular Hough transforms. The main challenge is to interpret finger feature points in the patterns of circles amidst noise. The approach was implemented in java and tested on a set of images. The results were assessed using manual visual inspection. Such spatial approaches hold potential for more robust and flexible hand related image analysis application. Moreover, these approaches could also give faster algorithms as there is no need for image binarization and threshold optimization.