An algorithm to Convert the Gestures of Numbers of Four languages into Voice

Abstract

Gestures are one of the best ways of communication between dumbs people they can speak and other people they can speak using the expression of signs language. It is good to communicate between the dumb and other people by convert their signals into voices to be easy communicate with other people they can speak and hear voices. In this paper, a new algorithm proposed for recognizing hand gestures of numbers (0-10) to four languages (English, Arabic, Chinese and Persian languages) depending on dumbs signs and convert the signs into voices corresponding to signs numbers of each language. The proposed algorithm, firstly uses video for gesture of the dumb to three languages then converts these videos into frames (images) , secondly preprocessing step to removing the noises, resizing the images and increasing the contrast, the third step is   extraction features step to calculating the distance of clustering algorithms such as Bayesian, C4.5, k-mean, k- medoid and artificial neural network. Eighteen features are calculated; eight features from Euclidean distance, eight features from slop, Area, and perimeter.  The results in the training stage were; Bayesian gave 100% accuracy, C4.5 gave 100% accuracy, k-mean gave 100% accuracy k-medoid gave 100% accuracy and artificial neural network gave 95% accuracy. While in the testing stage classifiers are; Euclidian Distance, Modify Standardize Euclidian Distance and Correlation to calculating the difference between the features stored from training stage with new tested features and the results show that Euclidian Distance gave 100% accuracy, modified Standardize Euclidian Distance gave 100% accuracy and Correlation gave 100% accuracy. The database is created in our laboratory (six videos with 324 frames).