CONTOUR DESCRIPTIONS FOR MACHINE VISION

NORMALIZED INTERVAL/VERTEX DESCRIPTORS (NI/VD) ALLOW A SIMPLE, YET EXACT REPRESENTATION OF A CONTOUR BY CODING THE LENGTH OF STRAIGHT LINE SEGMENTS AND THE ANGLE CONNECTING THESE SEGMENTS INTO A CHAIN CODE. BY NORMALIZING THE LENGTHS THE REPRESENTATION IS MADE SCALE INVARIANT. BY CHOICE OF A STARTING SEGMENT, IT IS ROTATIONALLY INVARIANT. THEREFORE A "LIBRARY" REPRESENTATION OF THE CONTOUR IS COMPACT AND ROBUST OVER TWO IMPORTANT SOURCES OF CLASSIFICATION ERROR. PARTIAL CONTOURS ARE ACCURATELY REPRESENTED (TO A SCALING ERROR) BY THE LIBRARY. THE NI/VD IS USEFUL IN DEFINING A "COMPLEXITY MEASURE" OF THE CONTOUR, DEFINED AS THE SUM OF THE MAGNITUDE OF THE ANGLES BETWEEN SEGMENTS. CONTOURS AND LIBRARY DESCRIPTORS CAN BE CLASSIFIED SIMPLY BY THEIR COMPLEXITY. THEREFORE, CONTOURS MAY BE QUICKLY SORTEDINTO CLASSES OF COMPARABLE COMPLEXITY, INCREASING THE EFFICIENCY OF CONTOUR IDENTIFICATION. SEGMENTATION OF IMAGES MAY PROCEED FROM A VARIETY OF PRE-PROCESSING ALGORITHMS. SCENE TYPES ARE APPROPRIATELY SEGMENTED BY CHANGES IN INTENSITY, DIFFERENCES IN TEXTURE, OR BY COLOR BOUNDARIES. NO MATTER WHICH PREPROCESSING ALGORITHM IS APPLIED, THE RESULT IS A COLLECTION OF CONTOUR LINES REPRESENTING THE RESULTS OF THE SEGMENTATION PROCESS, UPON WHICH THE NI/VD IS APPLIED.