Distillation of an End-to-End Oracle for Face Verification and Recognition Sensors.

Face recognition functions are today exploited through biometric sensors in many applications, from extended security systems to inclusion devices; deep neural network methods are reaching in this field stunning performances. The main limitation of the deep learning approach is an inconvenient relation between the accuracy of the results and the needed computing power. When a personal device is employed, in particular, many algorithms require a cloud computing approach to achieve the expected performances; other algorithms adopt models that are simple by design. A third viable option consists of model (oracle) distillation. This is the most intriguing among the compression techniques since it permits to devise of the minimal structure that will enforce the same I/O relation as the original model. In this paper, a distillation technique is applied to a complex model, enabling the introduction of fast state-of-the-art recognition capabilities on a low-end hardware face recognition sensor module. Two distilled models are presented in this contribution: the former can be directly used in place of the original oracle, while the latter incarnates better the end-to-end approach, removing the need for a separate alignment procedure. The presented biometric systems are examined on the two problems of face verification and face recognition in an open set by using well-agreed training/testing methodologies and datasets.

High dynamic range image display with halo and clipping prevention.

The dynamic range of an image is defined as the ratio between the highest and the lowest luminance level. In a high dynamic range (HDR) image, this value exceeds the capabilities of conventional display devices; as a consequence, dedicated visualization techniques are required. In particular, it is possible to process an HDR image in order to reduce its dynamic range without producing a significant change in the visual sensation experienced by the observer. In this paper, we propose a dynamic range reduction algorithm that produces high-quality results with a low computational cost and a limited number of parameters. The algorithm belongs to the category of methods based upon the Retinex theory of vision and was specifically designed in order to prevent the formation of common artifacts, such as halos around the sharp edges and clipping of the highlights, that often affect methods of this kind. After a detailed analysis of the state of the art, we shall describe the method and compare the results and performance with those of two techniques recently proposed in the literature and one commercial software.