A Genetic Algorithm-Based 3D Feature Selection for Lip Reading

In lip reading, selection of features play crucial role. In lip reading applications database is video, so 3 Dimensional transformation is appropriate to extract lip motion information. State of art the lip reading is based on frame normalization and frame wise feature extraction. However this is not appropriate due to chances of information loss during frame normalization. Also all the frames cannot be considered equally as they bear varying motion information. In this paper 3D transform based method is proposed for feature extraction. These features are the input to Genetic Algorithm (GA) model for discriminative analysis. Genetic Algorithm is used for dimensionality reduction and to improve the performance of the classifiers at low cost of computation. Both testing and training time for classifier is reduced by compact feature size. For experimentation of digit utterances CUAVE and Tulips database are used. The results obtained are compared with various feature selectors from WEKA software. It is found that from classification accuracy point of view proposed method is better than others.