Sign Languages (SLs) are the basic means of communication between deaf people all over the world. Sign Language recognition systems usually focus on recognizing sequences of hand morphs constituting signs. Recognizing hand morphs of Greek Sign Language (GSL) letters is the basis of the recognition system that we have developed. Visual-based systems have been used by many researchers, to increase naturalness in user's movement. In such systems images are captured by a camera, in a time sequential basis and they are used for training and recognition. Most times the acquired image cannot directly be used for recognition purposes. Some preprocessing is required in order to extract useful information from images. It is a crucial step, however, to select good features in any object recognition system. In our approach we obtain hand morph images that we capture with the use of a monochrome video camera. We then process each hand morph image, representing a GSL letter, by applying filtering methods on it, in order to get a much simpler form of the image. We then extract a feature vector from each image, by calculating geometrical properties of the hand morph. The elements of the feature vectors are the lengths of vectors, that originate from the center of mass and end up to the fingertips area. The fingertips area is the one that bares important information for each hand morph. We tried several feature vectors of various lengths, in order to select the optimum vector. Experimental results led us to the selection of a feature vector of 37 elements. The specific vector has the advantage of small size, which improves system's performance. The kind of vector is descriptive for the hand morph, and proved to be effective, too. For all these reasons it has been considered appropriate for the recognition of sequences of letters, too. The system should be able to recognize various instances of hand morph letters, so as to be capable of recognizing hand morphs of different persons. For this reason various instances of hand morphs are used in the training and test phases for each letter. This paper summarizes our analysis and experimental results for the performance evaluation of a single letter recognition system for GSL alphabet. The development of such a recognition system proves the effectiveness of the specific feature vector that we have selected.
Cite as: Pashaloudi, V.N., Margaritis, K.G. (2004) A performance study of a recognition system for Greek sign language alphabet letters. Proc. 9th Conference on Speech and Computer (SPECOM 2004), 545-551
@inproceedings{pashaloudi04_specom,
author={Vassilia N. Pashaloudi and Konstantinos G. Margaritis},
title={{A performance study of a recognition system for Greek sign language alphabet letters}},
year=2004,
booktitle={Proc. 9th Conference on Speech and Computer (SPECOM 2004)},
pages={545--551}
}