Employing Kaze Features for the Purpose of Emotion Recognition

摘要

In this research, a novel approach for emotion detection is exploited by taking the Accelerated Kaze (A-Kaze) features for emotion recognition. The Kaze Features work in a way such that object boundaries can be preserved by making blurring locally adaptive to the image data without severely affecting the noise-reducing capability of the Gaussian blurring, thereby increasing the accuracy of the system. After extracting the Kaze features, GMM is constructed and thus a Fisher Vector representation is made. The extracted features are passed through an SVM detector. An efficiency of 87.5% has been shown thus proving that Kaze can also be used effectively in the field of facial image processing.