Introduction to the Special Issue on Learning in Computer Vision and Pattern Recognition

THE GOAL OF computer vision and pattern recognition (CVPR) research is to provide computers with human-like perception capabilities so that they can sense the environment, understand the sensed data, identify patterns, take appropriate actions, and learn from this experience in order to enhance future performance. The field has evolved from the application of classical pattern recognition and image processing techniques to advanced applications of image understanding, dynamic scene analysis, model-based vision, knowledge-based vision and systems that exhibit learning capability. Over the years, there has been an increased demand for CVPR systems to address “real-world” applications, such as autonomous navigation, target recognition, manufacturing, photointerpretation, remote sensing, situation awareness, image/video database management, etc. This requires that the vision techniques be robust and flexible to optimize performance in diverse scenarios encountered in a given application. Past research in applying learning techniques to CVPR problems has been limited [1], [2]. Some of the reasons for this were the lack of understanding and availability of tools for low-level image analysis, interdisciplinary nature of learning in CVPR research and the lack of machine learning [14]–[16] and statistical learning tools [12], [13]. However, in the last decade, some progress has been achieved toward these problems. Solving the signal-to-symbol transition problem remains one of the key challenges in the application of symbolic learning to vision. Learning requires a lot of data and speed for its practical use. The field of machine and statistical learning is driven by the idea that computer algorithms and systems can improve their own performance with time. Vision provides interesting and challenging problems and a rich environment to advance the state-of-the art in learning. There is a strong potential in learning technology to contribute to the development of flexible and robust vision algorithms, thus improving the performance of vision systems for practical use. Learning-based vision systems are expected to provide a higher level of competence and greater generality. Learning may allow using the experience gained in creating a vision system for one application domain to a vision system for another domain by acquiring and maintaining knowledge. Thus, an innovative integration of learning and CVPR techniques has the promise of advancing the field which will contribute to better understanding of complex images of real-world dynamic scenes. There is another benefit of incorporating a learning paradigm in the computational CVPR framework. To mature the laboratory-grown vision systems