Recognition, Tracking, and Optimisation

This special issue of the International Journal of Computer Vision contains eight selected contributions that showcase some of the most actively researched areas in Computer Vision, ranging from object recognition and identification, motion analysis and tracking, and optimisation. It also includes examples of real world applications where computer vision offers reliable quantitative solutions. Advances in fundamental methods, such as learning algorithm and feature representation, are essential in performing these tasks, as the papers in this special issue show. Yu et al. present a novel algorithm for free-hand sketch recognition in “Sketch-a-Net: a Deep Neural Network that Beats Humans”. A novel deep neural network architecture suitable for sketch recognition is proposed along with data augmentation strategies. The augmentation technique incorporates the temporal order of strokes and both local deformation of stroke splines at local level and large deformations of entire object at global level. A joint Bayesian fusion method is used to perform the ensemble. The proposed method is shown to outperform the earlier systems and is computationally more efficient. In their paper “Deep Perceptual Mapping for CrossModal Face Recognition”, Sarfraz and Stiefelhagen address the issue of thermal-to-visible face recognition. This paper provides a deep learning approach tomulti-modal face recognition by matching visible RGB images to infra-red images. The authors propose a method for learning a non-linear mapping of local image descriptors across modalities using deep neural network, which is trained with features cal-