Symbolic Construction of a 2-D Scale-Space Image
IEEE Transactions on Pattern Analysis and Machine Intelligence
| Hi-index | 0.00 |
IN THIS PAPER WE OUTLINE AN APPROACH TO INTERMEDIATE-LEVEL PROCESSING WHICH USES SYMBOLIC TOKENS AND RELATIONAL MEASURES BETWEEN THEM AS THE BASIS FOR GROUPING. THE FIRST PART OF THE PAPER DEVELOPS A COMPUTATIONAL FRAMEWORK FOR BOTTOM-UP IMAGE ABSTRACTION USING ZERO-, ONE-, AND TWO-DIMEN- SIONAL TOKENS, I.E. POINTS, LINES, AND AREAS; STARTING WITH THE GENERATION OF TOKENS FROM THE IMAGE, AND FOLLOWED BY THE CREATION OF A HIERARCHY OF LEVELS OF ABSTRACTION. EACH ABSTRACTION STEP FROM ONE LEVEL TO THE NEXT EITHER GROUPS SEVERAL TOKENS INTO A SINGLE ONE (E.G. REPLACING A SET OF PARALLEL LINES BY AN AREA-TOKEN), REDUCES TOKENS (E.G. REPLACING A LONG, THIN AREA BY A LINE-TOKEN), OR ELIMINATES LESS SIGNIFICANT TOKENS. WHILE GROUPING DEPENDS ON THE DENSITY OF TOKENS AND THEIR FEATURES LIKE CONTRAST AND COLOR, OFTEN THE MOST IMPORTANT PROPERTY IS THEIR GEOMETRIC CONFIGURATION. WAYS TO REDUCE THE COMPUTATIONAL COMPLEXITY OF THE GROUPING PROCESSES ARE STUDIED, AND IN SO DOING THE ADVANTAGES OF A SYMBOLIC APPROACH OVER OTHER LOW-LEVEL VISION METHODS ARE DEVELOPED. THE SECOND PORTION OF THIS PAPER DESCRIBES AN IMPLEMENTATION OF COLLINEAR GROUPING FOR STRAIGHT LINE SEGMENTS, WHICH IS DEVELOPED WITHIN THE GENERAL ABSTRACTION FRAMEWORK OF THIS PAPER. THE INITIAL LINE SEGMENTS ARE GENERATED FROM GRADIENTS AT THE POSITION OF LAPLACIAN ZERO-CROSSING CONTOURS. THE NEXT STEP IS REPEATED IN A HIERARCHICAL FASHION: A GRAPH