Facial Expression Recognition Based on Local Gabor Binary Pattern Feature and Use Discrete Cosine Transform to Reduce the Feature Dimension

In facial expression recognition, local Gabor binary pattern histogram features have been proved to be more localized features, but it's still can't describe the detailed local texture, if only use the histogram statistics to distinguish facial expression, the calculation complexity and feature dimension are also remain high. This paper proposed a method of facial expression recognition based on local Gabor binary pattern features and used discrete cosine transform (DCT) to reduce the feature dimension, this method can retain the image texture information effectively, and reduced the feature dimension at the same time. First, used multi-scale and multi orientation Gabor filters on the image, and combined it with local binary pattern to extract features, then used DCT to reduce the feature dimension. The proposed method is verified the validity through the experiment on JAFFE facial expression database.