Design and Analysis of American Sign Language Classifier Based on n-tuple Technique

Abstract

This work consists of three stages: the first stage is the collection of data (images) for 26 letters alphabet, which is 1040 images, 40 for each class (letter). The second stage is the Preprocessing (segmentation, filtering, crop, convert to a binary image and feature extraction) and the last stage is the classy one. Two filters have been used median and wiener2 to treatment two types of noise; the salt and pepper and Gaussian noise respectively. The classification has been satisfied by using an n-tuple classifier algorithm is 93.8462%, while it is 93.2692% with salt and pepper noise. In this context, the recognition is 94.2308% when the median filter is used for salt & Pepper noise, while, it is 93.8462% when to use a Wiener2 filter with Gaussian noise. The optimal tuple size is 4, and suitable training pattern is 40 per each class.