Random Forests Expression Recognition Algorithm Based On Sequence Features

As a very important way of information exchanging during interpersonalcommunication, facial expression is able to express a lot of information that text andvoice can’t express. The analysis and recognition of facial expression is very significant.Facial expression recognition has a broad prospect of application in electronic games,intelligent advertisement delivery, remote education, safe driving, clinical medicine,psychology research, intelligent monitoring, facial image synthesis and animation,intelligent Human-Computer Interaction and so on. In recent years, as the expressionrecognition has been applied in more and more areas, people have a betterunderstanding of the important role of it. Facial expression recognition technology hasbecome the research focus of multiple fields.Currently, facial expression recognition research focuses on the expression featureextraction and expression classification algorithm. This paper put forward a new waythat based on Active Appearance Model and combined with the facial expressionsequence feature extraction method of Lucas-Kanade optical tracking algorithm andexpression classification method of Random Forest. First, use the AAM model to locatethe feature points in the neutral images and then use LK optical tracking algorithm totrack the AAM facial feature points at the next images. Second, treat the displacementof the AAM feature points between the peak images and their corresponding neutralimages as facial features. SVM algorithm is adopted to train classifier to classify theFacial Expression Action Units. Finally, put the Facial Expression Action Units intoRandom Forest as its inputs to train the facial expression classifier and it can recognizeseven basic expressions.This method has done a lot of tests on Extended Cohn-Kanade facial expressionimage sequence database. The results shows that AAM model combined with LKtracking algorithm is more accurate and efficient than AAM alone in the facialexpression sequence feature extraction. Use the displacement of AAM feature pointsbetween the first and the last image of a sequence as the inputs. Then, use SVM toclassify the AU can achieve above ninety-eight percent recognition rate. Use Facial Expression Action Units as the expression features and Random Forest as the classifier,the recognition rate is ninety-seven point one percent. But, at the same conditionBayesian network can only achieve eighty-nine point three seven percent recognitionrate. Random Forest is more efficient than Bayesian network at train and recognize.Random Forest has a lot of improvements than widely used Bayesian network at thefacial expression recognition rate and the efficiency of the algorithm.