Facial expression detection using directional local descriptor
In recent years, facial expression recognition is considered one of the most important challenges in image processing and has found many applications. Nowadays, due to the so-called emotional relationship between humans and computers in the virtual world, the use of facial recognition methods has become very important. In this research, a new method for recognizing facial expressions is proposed. In order to avoid the limitations and maintain the simplicity and efficiency of traditional LBP, we propose a simple but efficient conceptual and computational texture descriptor, which is called Local Triple Directional Pattern (LDTP). The main advantage of the proposed descriptor over the existing ones is that it combines both concepts of LTP and LDP operators in a similar compact coding scheme, which provides more accurate and separable information. It has also been shown that using a combination of features instead of just one feature makes facial expression better and more robust. Therefore, the features of LDTP and XCSLBP are combined to obtain the final feature vector. Then, for more speed, the final feature vector is dimensionally reduced using the PCA algorithm. Now, this feature vector is given to the SVM, which has been previously trained by the training data, to finally identify one of the 7 facial expressions. The proposed method was examined from various aspects and compared with other existing methods. The accuracy of the proposed method for facial expressions of 7 classes is approx