We are very pleased to introduce this special section of the Transactions on Pattern Analysis and Machine Intelligence on perceptual organization in computer vision. We feel that perceptual organization is at the heart of computer vision. Perceptual organization can play a key role in many other visual processes, such as recognition, motion understanding, stereo, and tracking. If we can gain a deeper understanding of this topic, we can make progress on many fronts at once.
This special section grew out of the Third International Workshop on Perceptual Organization in Computer Vision (POCV), held in Vancouver during the summer of 2001. Many of the papers submitted to this issue were first presented there in preliminary form. We are indebted to all the participants of the workshop for creating a lively atmosphere that encouraged us to produce this special issue and, especially, to Sudeep Sarkar and Kim Boyer, who began this series of workshops. We are pleased to see that the fourth POCV is coming up, under the guidance of Anthony Hoogs and Gerard Medioni.
As witnessed by these workshops, perceptual organization has been attracting renewed interest over the last decade. As a consequence of this interest, our special section received a large number of submissions, 35 in all, exceeding our greatest hopes (or perhaps our greatest fears, as we scrambled to find about 100 reviewers for these papers). As a consequence of the large number of submissions, we ended up with too many papers for a single issue. Therefore, we will also publish a special section in the June issue of TPAMI, to include four more papers that we could not fit into this issue.
The special section begins with two papers that consider image models, which are essential tools for PO. The first, by J. August and S.W. Zucker, deals with a stochastic model for contours that goes beyond the usual smoothness dependent characterization. The second work, by O. Ben-Shahar and S.W. Zucker, deals with texture modeling as two-dimensional flow where two types of curvatures control the texture appearance. Both papers show that these enhanced models are indeed effective in measuring perceptual affinities (grouping cues) and yield better discrimination between the modeled entities (curves or texture patches) in difficult cases.
Several papers in the special section address the problem of integrating nonlocal information in grouping. T. Tuytelaars, A. Turina, and L. Van Gool propose a grouping algorithm which is able to use nonlocal similarity but still stay efficient and effective, relying on geometric affine structure and symmetry properties. S. Mahamud, L.R. Williams, K.K. Thornber, and K. Xu builds on previous work on finding saliency by some of the authors. They use this as the basis for finding smooth closed contours efficiently.
Some papers in the special section analyze foundational issues about how problems in perceptual organization should be approached. The next paper is one example: E.A. Engbers and A.W.M. Smeulders focus on formal characterization and suggest a set of desirable properties that should be satisfied by grouping processes.
In addition to the paper of O. Ben-Shahar and S.W. Zucker, A. Hoogs, R. Collins, R. Kaucic, and J. Mundy propose another approach to characterizing texture. They consider texture as a collection of piecewise constant image segments and propose using the properties of these segments as well as their typical change over the image as texture characteristics. This approach allows one to identify texture under 3D transformations and even to identify structures that are usually not considered as texture.
We are also pleased to see papers that are concerned with testing the validity and the efficiency of general perceptual organization tools in a variety of domains. E. Saund's paper focuses on the problem of finding meaningful groups in hand-drawn sketches and line art to facilitate their editing. E. Saund uses classic gestalt cues of good continuation and closure while examining how these constraints should be modified and integrated into an efficient search process in this domain. M. Nicolescu and G. Medioni show how to apply the well-known approach of tensor voting to the problem of detecting boundaries in motion flow. A. Almansa, A. Desolneux, and S. Vamech show how to use principles of perceptual grouping to find lines and their vanishing points and to determine which of these computed properties is significant in the image. We feel that these papers provide valuable insight by focusing on specific problems. This allows us to evaluate frameworks for grouping in a specific context and inspires new insights about perceptual organization.
A. Almansa, A. Desolneux, and S. Vamech make use of a framework for perceptual organization proposed in the next paper, by A. Desolneux, L. Moisan, and J.-M. Morel. They suggest a new statistical criterion for testing whether a cluster is meaningful and demonstrate it using typical PO tasks such as histogram-based clustering and multilevel clustering.
The problem of handling nonlocal grouping cues is also attacked in quite a different way by B. Fischer and J.M. Buhmann. They point out that many image properties are slowly changing within a cluster and suggest an approach based on manifold clustering, which they apply to contours and textures.
The work in this issue is too broad in scope to allow us to make a glib generalization about its significance. All we can say is that, we are excited by recent advances in our field and look forward to seeing work in perceptual organization have an increasing impact on computer vision.
We have many people to thank for helping us with this special section. We are very grateful to all the reviewers who helped. The reviewers are truly the ones who shaped the issue; we have tried only to act as moderators to their views. We sincerely thank all the authors who submitted to this issue both for their exciting work and their patience with us as we grappled with an unexpectedly large editing task. We thank the NEC Research Institute for their generous contributions to our workshop. We thank the TPAMI staff who have helped and advised us. Most of all, we would like to thank Hadas Heier for all of her hard work and efficiency in handling the logistics of this issue.
David W. Jacobs
D.W. Jacobs is with the Department of Computer Science, University of Maryland, College Park, MD 20742. E-mail: email@example.com.
M. Lindenbaum is with the Computer Science Department, Technion, Haifa 32000, Israel. E-mail: firstname.lastname@example.org.
For information on obtaining reprints of this article, please send e-mail to: email@example.com, and reference IEEECS Log Number 118216.
David W. Jacobs
received the BA degree from Yale University in 1982. From 1982 to 1985, he worked for Control Data Corporation on the development of database management systems and attended graduate school in computer science at New York University. From 1985 to 1992, he attended the Massachusetts Institute of Technology, where he received the MS and PhD degrees in computer science. From 1992 to 2002, he was at the NEC Research Institute, Princeton, New Jersey, where he was a senior research scientist. In 1998, he spent a sabbatical at the Royal Institute of Technology (KTH) in Stockholm. Since 2002, he has been an associate professor of computer science at the University of Maryland, College Park. His research has focused on human and computer vision, especially in the areas of object recognition and perceptual organization. He has also published papers in the areas of motion understanding, memory and learning, and computational geometry. He is an associate editor of the IEEE Transactions on Pattern Analysis and Machine Intelligence
. He and his coauthors received honorable mention for the best paper award at CVPR 2000. He is a member of the IEEE and the IEEE Computer Society.
received the BSc, MSc, and DSc degrees from the Department of Electrical Engineering at the Technion, Israel, in 1978, 1987, and 1990, respectively. From 1978 to 1985, he served in the Israeli Defense Forces. He did his postdoctoral work at the NTT Basic Research Labs in Tokyo, Japan, and since 1991, he has been with the Department of Computer Science, Technion. He was also a consultant to Hewlett-Packard Labs, Israel, and recently spent a sabbatical at the NEC Research Institute, Princeton, New Jersey (in 2001). He worked in digital geometry, computational robotics, learning theory, and various aspects of computer vision. Currently, his main research interest is computer vision and, especially, statistical analysis of object recognition and grouping processes. He is a member of the IEEE.