Analysis of Feature Invariance and Discrimination for Hand Images: Fourier Descriptors versus Momen

Abstract—This papers addresses the issue of feature extraction for the case of gesture recognition. Two important properties are covered in the experiments, invariance to certain transformations and discrimination power. The study of these two characteristics was carried out using Moment Invariants and Fourier Descriptors, specifically for ASL (American Sign Language) images, and for that purpose a new dataset for the ASL images was created. Three types of images greyscale, binary and contours were used with the Moment Invariants. Two different Fourier Descriptors were compared, namely complex coordinates and central distances. The results showed trends for the invariance and discrimination powers for both feature sets. The paper shows how to gather information that allow researchers to choose between two competing feature extraction methods before attempting to train a classifier.