Research On Bag Of Words Model-Based Facial Expression Recognition

In recent years, with rapid development and popularization of smart phones and othercomputing devices with high-definition camera, identification recognition application andother intelligent human-computer interactive applications based on facial image appearmore and more in people’s daily lives, because of the in-depth research and developmenton recognition of face features. Facial expression recognition is an important direction inthe face characteristics recognition technology, there are good prospects for developmentin the increasingly intelligent application scenarios, and it’s a popular research branchnow.Bag of words model has been proposed to be used in image-based scene recognition,the principle is to descript objects distribution in the scene with disordered visualvocabulary collection, in order to distinguish between different scenes. Unlike scenepicture, face components in a face image of human are with high stability andconsistency, and so disordered visual vocabulary collections gained out of the bag ofwords model directly applying to the whole face image for facial expression recognition,are difficult to provide a high enough degree of expression discriminations. Sub-block orcomponent-based bag of words model is proposed and applied to facial expressionrecognition. This method uses bag of words model to generate several unorderedcollections of visual vocabulary, in order to extract local features with strongerdiscriminations, and then concatenate these visual vocabulary collections into a large onefor expression classification. However, the recognition rate of this method is not highenough, because the visual vocabulary of bag of words model is unordered, which itselfignores the spatial distribution of the image feature, resulting in the local features of largerregion having a higher degree of discriminations are separated into a number of smallerlocal features that seprate from each other, and these features don’t have enoughdiscriminations for expressions. Therefore, we add spatial pyramid matching to thecomponent-based bag of words model, and this improved recognition method can extractlocal features with stronger discriminations than the the visual vocabulary collections ofthe bag of words model, thereby enhancing the expression recognition rate. We did facial expression recognition experiments with component-based bag ofwords model and additional space pyramid matching improved method component-basedbag of words model and additional spatial pyramid matching improved method onCohn-Kanade face image database. The improved method obtained a more satisfactoryrecognition rate, and relative to the original method, the recognition rate is greatlyimproved.