Face Attribute Recognition Based On Tree Structure

In recent years, with the progress of science and technology, camera updating fast,computer and Internet technology has developed rapidly, human beings have entered the era of big data. As an assist of text, digital images with their better intuition record the intravenous drip of humans’ life. With a large number of images, how to make the machine can imitate people’s cognitive habits. That is to say, machines can use prior semantic description(attribute) knowledge to learn some similar things unknown. To solve this problem, this paper use face images as a representation to do research from two aspects.As we all know, the facial expression is common through race, gender, age,education level, color and other factors. And the concentrated areas of facial expression information are eyebrows, eyes, mouth and so on. For this, we put forward the facial expression intensity measurement analysis method based on decision tree.First, for a given training sample, we construct the facial expression information model by analyzing the characteristic points. Then according to humans’ understanding and description experience of facial expression, I analyze characteristic points parameters, design to generate the decision tree to measure facial expression intensity. Finally, according to the facial expression intensity measurement of the decision tree, classify the test samples.Aiming at learning real life faces’ attributes with complex background, we build a human face attribute identification system based on the mixture of trees. First of all,build a original face mixture of trees model. Then given training samples, the training samples contain positive training samples with human faces and negative training samples without human faces, and label the characteristic points of positive training samples, using the training samples through Discriminative Training method to get new faces attribute mixture of trees model. Finally, combining with the face detection method, using the learned mixture of trees model to get the attribute information of detected human faces.In the experiment, three face databases(Cohn-Kanade database,Public Figure Face Database, CMU Multi PIE database), a non-face database(INRIAPerson database) and images contain human face in real life are used to make research for smiling intensity analysis and face attribute information learning in complex images. Through the experiments, we can see that trees structure plays a useful role for attribute learningand relative analysis.