Handcrafted 2016 - Special Issue on Handcrafted vs. Learned Representations for Human Action Recognition

Human action recognition has long been one of the most active topics in computer vision in the last few decades. Its applications can be found in many important areas such as video surveillance, video annotation and retrieval and human-computer interaction. Action representations (including both local and holistic representations) play a fundamental role in action recognition. Conventional methods mainly rely on low-level handcrafted features by feature engineering, e.g., histogram of three-dimensional oriented gradients (HOG3D), histogram of oriented gradients and histogram of optical flow (HOGHOF) and spatial-temporal oriented energies. Recently, feature learning has drawn increasing interest in visual recognition, especially, for scene classification and digit recognition in the image domain. Both feature learning and feature engineering have their advantages in visual representation which, however, remains less explored for action recognition. It is therefore urgent and highly desirable to innovate advanced representation techniques for human action recognition and boost the progress in this field towards practical applications.
Feature engineering is able to incorporate human ingenuity and prior knowledge and is still widely used in both image and video domains. Compared to feature learning, feature engineering enjoys the flexibility and computational efficiency, and does not rely on large sets of samples for training. Feature learning, e.g., deep learning, can directly learn data representations from raw training samples and detect data-driven features for specific tasks. In contrast to feature engineering, feature learning can extract and organize the discriminative information from the data and novel applications can be constructed faster. In addition, feature learning algorithms, e.g., descriptor learning, can also construct objective functions by parameterization of handcrafted features to learn task-specific descriptors. Up until now, feature representation for action recognition is still a challenging task far from being solved. It is highly desirable to design efficient and effective action representations by both feature engineering and feature learning for human action recognition.
This special issue targets researchers of broad fields from diverse communities, including machine learning, computer vision and pattern recognition and will encourage novel theories and advanced techniques of feature learning/engineering for human action recognition. This special issue is devoted to the publications of high quality papers on technical developments and practical applications around feature learning/engineering for both local and holistic representations of human actions. It will serve as a forum for recent advances in the fields of multimedia analysis, computer vision, machine learning, etc. We invite original and high quality submissions addressing all aspects of these fields. Relevant topics include, but are not limited to:
Efficient feature engineering and descriptor design algorithms for action representation
Feature learning for both local and global representations of human actions
Discriminative/supervised descriptor learning for action recognition
Manifold/subspace learning for action recognition
Evolutionary feature learning for action recognition
Domain adaption for action recognition
Efficient deep learning for action representation
Sparse representation learning for action recognition
Efficient feature/descriptor learning algorithms for action recognition
Manuscript Submission: Manuscripts should be formatted and be submitted online according to the instructions for Image and Vision Computing at http://www.elsevier.com/journals/image-and-vision-.... The authors must select “SI: Handcrafted vs. Learned” when specifying the “Article Type” in the submission system. Submitted manuscripts must not have appeared or been under review elsewhere. All submitted manuscripts will be reviewed by at least three reviewers in accordance with the refereeing procedure of Image and Vision Computing, and only those manuscripts that require minor revisions will be accepted for rapid publication in this special issue.
Important Dates:
Submission Deadline: 15 August 2015
Acceptance/Revision Notification: 1 November 2015
Final Decision: 15 March 2016
Tentative Publication: 2016
Guest Editors:
Dr. Xiantong Zhen
The University of Western Ontario, London, ON, Canada.
Email:zhenxt-AT-gmail.com
Webpage: http://www.digitalimaginggroup.ca/members/xtzhen.p...
Prof. Ling Shao
Northumbria University, Newcastle upon Tyne, United Kingdom.
Email: ling.shao-AT-ieee.org
Webpage: http://lshao.staff.shef.ac.uk/
Prof. Steve Maybank
Birkbeck College, University of London, London, United Kingdom.
Email: sjmaybank-AT-dcs.bbk.ac.uk
Webpage: http://www.dcs.bbk.ac.uk/~sjmaybank/
Prof. Rama Chellappa
University of Maryland, College Park, MD, USA
Email: rama-AT-umiacs.umd.edu
Webpage: http://www.umiacs.umd.edu/~rama/