Deep Learning methods are able to automatically discover better representations of the data to improve the performance of the classifiers. However, in computer vision tasks, such us the gender recognition problem, sometimes it is difficult to directly learn from the entire image. In this work we propose a new model called Local Deep Neural Network (Local-DNN), which is based on two key concepts: local features and deep architectures. The model learns from small overlapping regions in the visual field using discriminative feed-forward networks with several layers. We evaluate our approach on two well-known gender benchmarks, showing that our Local-DNN outperforms other deep learning methods also evaluated and obtains state-of-the-art results in both benchmarks.

Abstract. In this work, we have studied the viability of a novel technique to estimate the POR that only requires video feed from a consumer camera. The system can work under uncontrolled light conditions and does not require any complex hardware setup. To that end we propose a system that uses PCA feature extraction from the eyes region followed by non-linear regression. We evaluated three state of the art non-linear regression algorithms. In the study, we also compared the performance using a high quality webcam versus a Kinect sensor. We found, that despite the relatively low quality of the Kinect images it achieves similar performance compared to the high quality camera. These results show that the proposed approach could be extended to estimate POR in a completely non-intrusive way.

In the present work, we propose to deal with two important issues regarding to the RBM's learning capabilities. First, the topology of the input space, and second, the sparseness of the RBM obtained. One problem of RBMs is that they do not take advantage of the topology of the input space. In order to alleviate this lack, we propose to use a surrogate of the mutual information of the input representation space to build a set of binary masks. This approach is general and not only applicable to images, thus it can be extended to other layers in the standard layer-by-layer unsupervised learning. On the other hand, we propose a selective application of two different regularization terms, L 1 and L 2 , in order to ensure the sparseness of the representation and the generalization capabilities. Additionally, another interesting capability of our approach is the adaptation of the topology of the network during the learning phase by means of selecting the best set of binary masks that fit the current weights configuration.The performance of these new ideas is assessed with a set of experiments on different well-known corpus.

Abstract. This paper deals with automatic feature learning using Gaussian Restricted Boltzmann Machines (GRBM) for the problem of gender recognition in face images. The GRBM is presented together with some practical learning tricks to improve the learning capabilities and speedup the training process. The performance of the features obtained is compared against several linear methods using the same dataset and the same evaluation protocol. The results show a classification accuracy improvement compared with classical linear projection methods. Moreover, in order to increase even more the classification accuracy, we have run some experiments where an SVM is fed with the non-linear mapping obtained by the GRBM in a tandem configuration.