Kennedy Chengeta¹ and Serestina Viriri², ¹University of KwaZulu Natal, South Africa and ²Westville Campus, South Africa

Facial expression recognition in the field of computer vision and texture synthesis is in two forms namely static image analysis and dynamic video textures. The former involves 2D image texture synthesis and the latter dynamic textures where video sequences are extended into the temporal domain taking into account motion. The spatial domain texture involves image textures comparable to the actual texture and the dynamic texture synthesis involves videos which are given dynamic textures extended in a spatial or temporal domain. Facial actions cause local appearance changes over time, and thus dynamic texture descriptors should inherently be more suitable for facial action detection than their static variants. A video sequence is defined as a spatial temporal collection of texture in the temporal domain where dynamic features are extracted. The paper uses LBP-TOP which is a Local Binary Pattern variant to extract facial expression features from a sequence of video datasets. Gabor Filters are also applied to the feature extraction method. Volume Local Binary Patterns are then used to combine the texture, motion and appearance. A tracker was used to locate the facial image as a point in the deformation space. VLBP and LBP-TOP clearly outperformed the earlier approaches due to inclusion of local processing, robustness to monotonic gray-scale changes, and simple computation. The study used Facial Expressions and Emotions Database(FEED) and CK+ databases. The study for the LBP -TOP and LGBP-TOP achieved bettered percentage recognition rate compared to the static image local binary pattern with a set of 333 sequences from the Cohn–Kanade database.

Local binary patterns on Three Orthogonal Planes (LBPTOP) · Volume Local Binary Patterns(VLBP)