Recognition between smiling and neutral facial display with power LBP operator

Automatic recognition of emotions can be helpful in many situations. Therefore, research towards providing tools enabling the most accurate recognition is necessary. In this work, the problem of smiling and neutral facial display classification is addressed. For image description, the local binary patterns (LBP) as well as the novel power LBP (PLBP) texture operators are exploited. Their performance is compared on several databases and examined for variations from the standard designs, such as uniform and rotation invariant approaches. The classification is performed with the use of support vector machine, SVM. The obtained results show that when applying detailed image division schema, which is characterised by many small image patches analyzed separately, the introduced PLBP overcomes the LBP method. Moreover, the uniform or rotation invariant LBP versions do not improve the classification. However, the accuracy achieved using different types of PLBP is higher than when applying the standard LBP technique.