Facial expression recognition using local Gabor gradient code-horizontal diagonal descriptor

In this paper, we propose a novel method for facial expression recognition based on the combination of using Gabor filters and Local Gradient Code-Horizontal Diagonal (LGC-HD) features to generate the Local Gabor Gradient Code-Horizontal Diagonal (LGGC-HD) descriptor. In this approach, we firstly extract the Gabor Features Map (GFM) by convolving the facial image with three scales and eight orientations of the Kernel Gabor Filters (KGFs). We then apply the LGC-HD on each image in the GFM in order to increase the gradient information features to obtain the LGGC-HD descriptor. Histogram sequence concatenation is then applied to the LGGC-HD descriptor to represent the facial image after dividing it into a number of nonoverlapping blocks. Finally, the Chi-square distance is employed to classify the different emotions. To increase the recognition rate and reduce the classifier time, a weighted mask is adopted, in which specific regions in the face were labelled such as eye, eyebrow and mouth with higher weights than other regions. Comparison results demonstrate the efficiency of the proposed method against other methods that utilized different approaches for describing the histogram sequence features such as Local Gabor Binary Pattern (LGBP). Experimental results on the JAFFE and CK databases showed high recognition rate of 93.33% and 90.62%, respectively.