Research On The Facial Expression Recognition Based On HMAX Model

Facial expression recognition is a comprehensive topic that relates to pattern recognition, image processing, artificial intelligence, etc. And it means that the compute is to analyze and detect the special expression state from the given expression image and then to ascertain the subject's specific emotion automaticly, and achieve an Intelligent interaction between human and computer. It has potential applications in many areas what includes robotics technology, image understanding, video retrieval, synthetic facial animation, psychology study, virtual-real technology, etc.Of facial expression recognition study consists mainly of three parts:face detection, facial feature extraction and expression classification. Many computer vision researchers had been done many studies on them, while there are still some problems not satisfactorily resolved, including the face false detection, facial recognition robustness, etc.Analyzing the technology and algorithms that relate to facial expression recognition at home and abroad, the thesis focuses on researching face detection, facial feature extraction and expression classification, and proposing solutions to problems which are mentioned above. The main content and the achievements of the research are as follows:1. Aspect of the face detectionThe thesis studies Viola's face detection method and face template matching method, analyzing the reason why Viola's face detection method produces face false detection, and then presents a method that combines Viola's face detection method with face template matching method to reduce the face false detection rate.The thesis searchs the suspicious face regions with Viola's face detection method firstly, and then uses face template matching technology to judge whether these regions include face really.2. Aspect of facial feature extractionResearching the distribution of expression feature and HMAX model, and then the thesis presents an improvement for feature blocks template extraction method on HMAX model according to the characteristics of distribution of expression feature, finally using HMAX model to extract the facial feature, which is called HMAX feature. The HMAX feature has a good invariance in scale, translation and direction. Compareing with other feature, which is extracted through other image processing methods, the HMAX feature has a better effect to distinguish different expression.3. Aspect of facial expression classificationThe thesis designes a facial expression recognition classifier based on AdaBoost.M2 learning algorithm and BP neural network.Using BP neural network as a weak classifier, the thesis uses AdaBoost.M2 learning algorithm to make a strong classifier as an expression recongition classifier.4. Based on the presented theory above, the thesis carry out experiments about face detection, facial feature extraction and expression classification to verify the correctness of the methods that are presented above.