Vowel Recognition by Using the Combination of Haar Wavelet and Neural Network

The lips movements are important in speech recognition and the Lip image segmentation has a significant role in image analysis. In this paper we present a novel technique to recognize Persian Vowels. The method is based on face detection and pupil location. First we perform the lip localization, then the color space CIE L*U*V* and CIE L*a*b* is used in order to improve the contrast between the lip and the other face regions. After that, the lip segmentation by using the Haar wavelet has done and the feature vectors has been extracted from the Haar wavelet result. Finally, the extracted feature vector has been used as neural network inputs and the vowels recognized. The proposed method has been applied on 100 tested images and the accuracy is about 79%.