Search

Identifying emotional sentiment projected in an image is a tedious task, considering the fact that sentiment represented by an image could depend on a very diverse set of factors. This paper presents a novel approach to predict the emotional sentiment of a group of people in a variety of environments. The proposed technique uses local facial features of subjects along with global scene features to estimate the type of emotional sentiment in group-level emotion recognition. Two separate convolutional neural networks based on different architectures are designed to predict group-level emotions into three categories: negative, neutral and positive. The first convolutional neural network referred as Scene-model, learns the global features in data. A novel partial fine-tuning process is proposed to train the model on task-specific data. The second convolutional model referred as Face-model is trained on facial expression datasets to learn the emotional status of subjects in an image. Joint distribution of the global (scene) and local (face) features is modeled using long short-term memory networks. This joint distribution is converted into class scores using softmax regression-based model. [ABSTRACT FROM AUTHOR]

Highlights • An activation function with data-range-segment-wise nonlinear behavior is proposed. • Values of slope coefficients for the proposed function are evaluated empirically. • Proposed activation function is presented in adaptive and non-adaptive variants. • Performance of proposed function is evaluated against state-of-the-art. • Internal activation behavior and convergence characteristics are analyzed. Abstract Activation functions play a crucial role in discriminative capabilities of the deep neural networks. They are also one of the main reasons for revival of the neural networks. Although, recent activation functions provide solution to vanishing and exploding gradient problems, but they do not have sufficient capacity to model non-linear data. This paper proposes a novel activation function, which imparts neural networks with capability to model non-linear dependencies in data. Behavior of proposed activation function remains non-saturating even with non-linear structure. The activation function provides distinct activation behaviors to different data range segments. The paper presents two different variants of proposed function. The first, "linear sigmoidal activation" function is a fixed structure activation function with function coefficients defined at the start of model design. Whereas second, "adaptive linear sigmoidal activation" function is a trainable function which can adapt itself according to the complexity of the given data. Both of the proposed models are tested against the state of the art activation functions on benchmark datasets (CIFAR-10, MNIST, SVHN, FER-2013). The proposed activation function is able to outperforms every known activation functions in all the tests. [ABSTRACT FROM AUTHOR]