Premium
Integrating artificial and psychophysical approaches for boundary finding in line drawings
Author(s) -
Tambouratzis T.
Publication year - 1995
Publication title -
international journal of intelligent systems
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.291
H-Index - 87
eISSN - 1098-111X
pISSN - 0884-8173
DOI - 10.1002/int.4550100502
Subject(s) - boundary (topology) , object (grammar) , segmentation , computer science , artificial intelligence , line (geometry) , boundary line , line segment , computer vision , process (computing) , illusory contours , image segmentation , pattern recognition (psychology) , geometry , mathematics , perception , optical illusion , mathematical analysis , neuroscience , biology , operating system
A novel approach for accurate boundary finding in line drawings of blocks‐world scenes is presented. Boundary finding aims at uncovering the exterior, occluding lines of the depicted objects, i.e., those lines which correspond to edges between nonabutting faces and which thus segment the scene. Boundary finding also helps restrict the further processing leading to structural analysis and scene interpretation. the proposed approach combines Winston's (1984) artificial scene analysis boundary stroll and Walters' (1987) psychophysics‐based enhancement technique. This approach has already been implemented and tested, illustrating advantages over both constituent techniques. the following main results are of special interest: (1) Exactly those lines in the line drawing which represent the segmentation edges of the solids in the corresponding scene are uncovered, irrespective of both the number and configuration of the viewed solids. (2) the inside/outside relations of the regions separated by the detected boundary lines are correctly determined, due to the direction‐sensitive nature of boundary finding. (3) Segmentation of the line drawing into separate object‐structures is automatically performed. All the sets of boundary lines uncovered by the proposed boundary finding approach constitute closed contours, whereas each object‐structure is the result of a region‐grouping process of the inside regions of an uncovered closed contour. (4) the general characteristics of each of the segmented object‐structures in the line drawing are exposed. (5) Parallelism is introduced as the scene is partitioned into segmented object‐structures on which simultaneous but independent application of scene analysis can subsequently be performed. © 1995 John Wiley & Sons, Inc.