Research On The Smile Recognition Based On Feature Fusion

Smile recognition has recently attracted significant attention due to the increasingaccessibility of inexpensive digital products and the development of pattern recognitiontechnology. Moreover, smile is regarded as an important expression, and theimprovement of smile recognition can promote the development of the expressionrecognition. As a result, improving the performance of smile recognition has become avaluable task. The main contents of this dissertation are as follows:1. The smile recognition based on the feature of face is studied. The LBP as thedescriptor for the face is selected. Given that the inherent characteristics of smileexpression, the face block-weighed LBP features for smile recognition are usedcombined with the precise of the face localization. The experimental results show thatthis method is suitable for the strong expressions.2. The smile recognition based on the feature of mouth is studied. The mouthlocalization and the description of the mouth feature are discussed in detail: theapproximate mouth region can be obtained based on the geometry features of face, andthen the accurate mouth region can be obtained by Ostu method combined with thehistogram specification; The HOG as the descriptor for the mouth is selected. Theexperimental results show that this method performs well compared to the first methodfor the smile expression.3. The smile recognition based on the feature fusion is studied. The researchabout the fusion algorithms based on series connection, canonical correlation analysis(CCA), and the discriminative CCA (DCCA) are discussed in detail. The experimentalresults demonstrate that the effectiveness and superiority of the fusion based on DCCA.4. The three feature representations are used to expression recognition. Given thatthe difference between the expression recognition and the smile recognition, the featureselection and the multi classification of the support vector machine (SVM) are analyzedchiefly. And the performance of expression recognition is analyzed in detail by theexperimental results.