Facial Emotion Recognition as Spatial Image using Gabor Filter

Abstract

The state of mind of a person can be easily understood from the human face. This paper proposes a methodology to recognize facial expression using Gabor filters, ResNet and two other custom models. The image is taken as the input data from the camera. This input can be used to extract information to infer a person's mood. First, we develop an algorithm for detecting image of an individual from entire set of images using Haar Cascade face detection algorithm. Then, we apply Gabor filter for extracting facial features in the spatial domain. Using the Gabor filter can effectively reduce computation and size, and in some situations even improve recognition. Gabor filters are used to capture the entire frequency spectrum in all directions. Finally, facial expressions are successfully classified by proposed Convolutional Neural Network model using extracted important facial features from the facial image after applying Gabor filter as input. The results of testing images from the CK+ dataset show the reliability and the best recognition rate of the proposed method.