Deep learning in digital marketing: brand detection and emotion recognition

Abstract

Deep learning has gained major popularity in automated feature extraction from images, audio and text. We present two case studies where deep learning can have a key impact. The first case study consists of a graphic logo detection based on a fast region-based convolutional networks (FRCN). This method tackles the issue of the logo different size and positioning by looking for scale invariant regions. This avoids a full image search while improving the overall object detection. Furthermore, instead of building a convolutional neural networks (CNN) from scratch, transfer learning and data augmentation techniques were applied excelling previous approaches. The second case study consists of a robust facial emotions recognition based on an improved version of the classic CNN-LeNet-5. Despite the net simplicity, it was found to be better suited for the system constraints, such as dataset dimension, face size and composition, achieving better performance than deeper networks such as GoogleNet and AlexNet.