Facial Emotion Recognition using Image Filtering Techniques

In the present era, facial emotion recognition is one of the rapidly growing research fields that have various applications like human-computer interaction, security, health care, and human resource management. Facial expressions represent person’s emotional state. In this paper, the Butterworth filter is utilized to filter the image before sending it to CNN architecture. Recognition accuracy of the model without using the Butterworth filter is $\mathbf{9 8. 0 5 \%}$. In the proposed model images of the CK+ dataset have been pre-processed using the Butterworth filter, and then the dataset has been trained using the CNN model, which comprises two convolution layers and two max pooling layers. Analyzing emotions in images using proposed CNN model we have achieved improved test accuracy for various orders and cut off frequencies of Butterworth filter such as for order 4,5 and 6 with cut off frequency 25 model achieved test accuracy of $99.35 \%, 100 \%$, and $99.35 \%$ respectively.