Lip Reading by Using 3-D Discrete Wavelet Transform with Dmey Wavelet

Abstract

Lip movement is an useful way to communicate with machines and it is extremely helpful in noisy environments. However, the recognition of lip motion is a difficult task since the region of interest (ROI) is nonlinear and noisy. In the proposed lip reading method we have used two stage feature extraction mechanism which is précised, discriminative and computation efficient. The first stage is to convert video frame data into 3 dimension space and the second stage trims down the raw information space by using 3 Dimension Discrete Wavelet Transform (DWT). These features are smaller in size to give rise a novel lip reading system. In addition to the novel feature extraction technique, we have also compared the performance of Back Propagation Neural Network (BPNN) and Support Vector Machine(SVM) classifier. CUAVE database and Tulips database are used for experimentation. Experimental results show that 3-D DWT feature mining is better than 2-D DWT. 3-D DWT with Dmey wavelet results are better than 3-D DWT Db4. Results of experimentation show that 3-D DWT-Dmey along with BNNN classifier outperforms SVM.