Domain Adaptation for Visual Understanding - Special Issue in Pattern Recognition
摘要截稿:
全文截稿: 2018-10-31
影响因子: 7.196
期刊难度:
CCF分类: B类
中科院JCR分区:
• 大类 : 计算机科学 - 1区
• 小类 : 计算机:人工智能 - 1区
• 小类 : 工程:电子与电气 - 1区
Overview
Visual understanding is a fundamental cognitive ability in humans which is essential for identifying objects/people and interacting in social space. This cognitive skill makes interaction with the environment extremely effortless and provides an evolutionary advantage to humans as a species. In our daily routines, we, humans, not only learn and apply knowledge for visual recognition, we also have intrinsic abilities of transferring knowledge between related visual tasks, i.e., if the new visual task is closely related to the previous learning, we can quickly transfer this knowledge to perform the new visual task. In developing machine learning based automated visual recognition algorithms, it is desired to utilize these capabilities to make the algorithms adaptable. Generally traditional algorithms, given some prior knowledge in a related visual recognition task, do not adapt to a new task and have to learn the new task from the beginning. These algorithms do not consider that the two visual tasks may be related and the knowledge gained in one may be used to learn the new task efficiently in lesser time. Domain adaptation for visual understanding is the area of research, which attempts to mimic this human behavior by transferring the knowledge learned in one or more source domains and use it for learning the related visual processing task in target domain. Recent advances in domain adaptation, particularly in co-training, transfer learning, and online learning have benefited the computer vision research significantly. For example, learning from high-resolution source domain images and transferring the knowledge to learning low-resolution target domain information has helped in building improved cross-resolution face recognition algorithms. This special issue will focus on the recent advances on domain adaptation for visual recognition. Topics of interest include but are not limited to:
- Novel algorithms for visual recognition using co-training, transfer learning, Online (incremental/decremental) learning, covariate shift, heterogeneous domain adaptation, dataset bias
- Domain adaptation in visual representation learning using deep learning, shared representation learning, multimodal learning, evolutionary computation-based domain adaptation algorithms
- Applications in computer vision such as object recognition, biometrics, hyper-spectral, surveillance, road transportation, and autonomous driving