Design of human face detection and recognition system along with speech synthesis subtitle

Abstract

In this paper, an efficient human face detection, recognition and Text To Speech (TTS) based wishing system is designed. The system design involves face detection using Viola-jones object detection algorithm, train the data base by finding the speeded up robust features (SURF)[9] features of detected faces, match of test features with train database features using Fast library for approximate Nearest Neighbors(FLANN) based matching technique, and finally announcing recognized human name based on training using text to speech based speech synthesis. The main problem in this type of systems design is the test face may have difference in view point or scale or illumination, when compared with trained faces, these factors will affect the recognition accuracy. To improve the recognition accuracy one way is train the database with more number of faces with different viewpoints, different scales and with different illumination levels. This will lead to large database size. To overcome this problem in this paper, a new approach is implemented i.e. in the time database training, first the face in the input image is detected and crop it and trained it with SURF features, because of this, size of the database is reduced 75%. One big advantage in SURF features is those are scale invariant, rotate invariant, and in this paper the surf features are illuminations invariant because when the time of feature description time the extracted numbers are normalized to unity. And finally recognized human name is announced using Letter to Sound (LTS) type speech synthesis technique.