Browse Prior Art Database

THE CVAP PROJECT PROGRESS.REPORT 1984

IP.com Disclosure Number: IPCOM000128267D
Original Publication Date: 1985-Dec-31
Included in the Prior Art Database: 2005-Sep-15
Document File: 20 page(s) / 66K

Publishing Venue

Software Patent Institute

Related People

The Integral Project: AUTHOR [+3]

Abstract

This report briefly describes the research carried out during 1964 at the Computer Vision and Associative Pattern Processing Laboratory (CVAP) of the Department of Numerical Analysis and Computing Science at the Royal Institute of Technology in Stockholm (KTH).

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 6% of the total text.

Page 1 of 20

THIS DOCUMENT IS AN APPROXIMATE REPRESENTATION OF THE ORIGINAL.

THE CVAP PROJECT PROGRESS.REPORT 1984

TRITA-NA-8509 INTEGRAL-43

Preface

This report briefly describes the research carried out during 1964 at the Computer Vision and Associative Pattern Processing Laboratory (CVAP) of the Department of Numerical Analysis and Computing Science at the Royal Institute of Technology in Stockholm (KTH).

The contributions of the National Swedish Board for Technical Develop-ment (STU) and the Swedish Board for Space Activities (DFR) are grate-fully acknowledged. STU has provided the major part of the funding of the laboratory since its start in 1982. We also gratefully acknowledge the support of the National Research Council and the Wallenberg Founda-tion in the development of the Computer Vision and Graphics Laboratory at KTH. This laboratory jointly serves the CVAP group and the Department of Physics IV.

The CVAP research group has evolved from the so called Integral project at the department. Current research focusses on low and intermediate computer vision and on the study of associative pattern processing. Modelling and interaction aspects on computers graphics are also being studied from these viewpoints.

In the following sections we give an overwiew of the research being done during 1984 in these areas.

>An assumption has to be made: many (more than two) lines will generally not be concurrent by accident. Barnard (10) suggests that the search for concurrency is carried out by mapping the image plane to the Gaussian sphere. The reason for this is that the search should be carried out in a compact space. In fact, a similar argument was put forward by Duda and Hart (6) when they explored the Hough transform. They did not compactify the space, but they chose the polar equation for the line and hence bounded one of the two variables.

Other problems in a similar vein concern the determination of symmetries (see e.g. (11)) and occluding contours, (12), (13), both being of paramount importance in recognition tasks. All these problems deal with issues treated in classical projective geometry. Bits and pieces of the wealth of results in geometry have also been used in computer vision. However, we contend that the approaches taken have often been somewhat arbitrary. The basic methodological principle of geometry as demonstrated by Klein e.g. in (14) states that problems should be solved by "changing the background". In a terminology more familiar to practitioners in AI this implies that the key issue is representation. But representation is not only a question about choosing the primitives. It is also a matter of posing the problem in its right context, that is defining the solution space correctly. If we consider the problems on recovery of scene structure listed above, we observe that they all deal with projective notions. The proper context to pose the problems is therefore in terms of projective geometry. We shall show that this indeed can...