REAL TIME EMOTION RECOGNITION AND GENDER CLASSIFICATION

Abstract: The proposed prototype aims to classify the gender and the emotions of a person in real time or using the image of the person in live cam or hard copy of the photograph. The gender classification would be implemented by simple yet robust real time convolutional neural network. Whereas for the emotion recognition, instead of fully connected layers, this model boasts and consists of depth wise separable convolution. It not only reduces the number of parameters and computation utilized in convolution but further increasing the efficiency. It has proved to achieve success in image classification models in terms of both, i) in obtaining better models than previously possible for a given parameter count required to perform at a given level and ii) acquire state-of-the-art results. The output is provided in the form of classes and these are seven classes of emotion recognition (angry, fear, sad, happy, surprise, neutral, disgust) and two classes of gender classification (Male and Female). The present accuracy for gender classification is 95%, whereas the accuracy for facial emotion recognition is around 67%. Also, a large reduction of hyper parameters is the main goal to reduce the model size.