REALTIME EMOTION BASED MUSIC PLAYER USING IMAGE PROCESSING

Computer engineering

This paper implements the use of Haar Cascade Classifier for face detection and Convolutional neural network for the emotion detection and thereby playing a song accordingly. In order to obtain minimal processing, multilayer perceptron are implemented by CNNs. In comparison to various algorithms for image classification, CNNs observed to have little-processing. This implies that the filters used in CNNs are advantageous when compared to traditional algorithm. The model includes two convolutional neural networks (CNN) models: a five-layer model and a global average pooling (GAP) model. We combined these CNN models with transfer-learning models. For our transfer-learning models, we used three pre-trained models: ResNet50; SeNet50; VGG16. Our results are comparable with the state-of-the-art models; however, our models are more efficient in performance. The visualization of features directly can be less informative. Hence, we use the training procedure of back-propagation to activate the filters for better visualization. The multiple actions such as capturing, detecting the emotion and classifying the same can all be confined as one step through the use of CNN. The slow performances of the real-time approaches could be enhanced by regularizing the methods and by visualizing the hidden features. Hence the proposed approach could enhance the accuracy and the computation speed..

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.