Recognition of Mexican sign Language from frames in video sequences

Abstract

The development of vision systems capable to extracting discriminative features that enhance the generalization power of a classifier is still a very challenging problem. In this paper, is presented a methodology to improve the classification performance of Mexican Sign Language (MSL). The proposed method explores some frames in video sequences for each sign. 743 features were extracted from these frames, and a genetic algorithm is employed to select a subset of sensitive features by removing the irrelevant features. The genetic algorithm permits to obtain the most discriminative features. Support Vector Machines (SVM) are used to classify signs based on these features. The experiments show that the proposed method can be successfully used to recognize the MSL with accuracy results individually above 97 % on average. The proposed feature extraction methodology and the GA used to extract the most discriminative features is a promising method to facilitate the communication of deaf people.