Vowel recognition based on face images using fisher linear discriminant analysis

Abstract

Speech and voice recognition has a wide range of uses across industries, including embedded devices such as in smartphones, dictation and assistance applications, smart cars, and others. The input for a speech recognition system could be in the form of audio signals or visual images. This paper presents a vowel recognition system, as parts of a speech recognition system, from face images using Fisher Linear Discriminant Analysis (FLDA) method. Images of human faces are used as input for the system. The vowel recognition process consists of the Canny edge detection stage for ROI extraction, the FLDA method for feature extraction, and the Euclidean distance calculation for vowel classification. The output of the system is a written vowel character. The experimental results showed that the average success rate for the vowel recognition was 66%, with vowel "i" and "e" achieved 100% recognition accuracies.