GRADIENT DIRECTIONAL EDGE CODING (GDEC) FOR EXPRESSION RECOGNITION FROM FACIAL IMAGES

Abstract

Facial Expression Recognition (FER) requires an effective Expression descriptor that can provide sufficient discrimination between different facial expressions. However, the existing FER methods are susceptible to noise, distortions and not able discriminate flat regions from noisy regions. Hence, this paper proposes a new Face descriptor called as Gradient Directional Edge Coding (GDEC) which encodes the expression components through the directional edges. Initially, GEDC finds the gradients for each pixel and then encodes them with their neighbor pixel’s support.  The support is assessed based on the deviation in the direction of corresponding neighbor pixel with the mean direction of a local region. Each pixel is encoded a 7-bit code word among which the six bits are belongs to the directions of neighbor pixels and one bit is sign bit. After describing the expression, the classification is accomplished through Support Vector Machine at different kernels.  Experimental validation on Standard CK+ dataset shows an accuracy of 94.6300% which is outstanding compared to the state-of-the-art methods.