Vowel Recognition System by Lip-Reading Method Using Active Contour Models and its Hardware Realization

Even in noisy environments such as in machinery factories or in crushes, we sometimes wish to control equipments by using human voices. Although word recognition methods have been successfully developed, noises in environments cause serious recognition problems. In such cases, lip shape movements are expected to be useful as the supporting data to improve the performance of recognitions. In this paper, a method to extract outer and inner lip shapes from input face images by using active contour models is proposed. The extracted lip shape data are utilized to control equipments by lip-readings in noisy circumstances. Normalization method to reduce the effect of the lip size change depending on the distance from a camera to human faces and to improve the recognition rate is also mentioned. A three-hierarchy neural network is used for the recognition method. The proposed method is implemented as hardware circuits in a FPGA chip to process NTSC video signals in real-time. Experimental results of vowel recognition are shown to confirm effectiveness of the proposed method