Boosting Facial Expression Recognition in a Noisy Environment Using LDSP-Local Distinctive Star Pattern

This paper presents a local feature descriptor, the Local Distinctive Star Pattern (LDSP), for facial expression recognition. The feature is obtained from a local 3x3 pixels area by computing the directional edge response value. Each pixel is represented by two 4-bit binary patterns, which is named as LDSP feature for that pixel. Each face is divided into 81 equal sized blocks and histogram of LDSP codes from those blocks are concatenated to build the feature vector for classification. The recognition performance of our descriptor is tested on popular JAFFE dataset with Support vector Machine (SVM) as classifier. Extensive experimental results with prototype expressions show that proposed LDSP descriptor is superior to existing appearancebased feature descriptor in terms of classification accuracy on static image.