Control point algorithms for contour processing and shape analysis
Planar non-overlapping shape analysis is an important computer vision problem with applications in industrial parts inspection, character recognition, target identification, biological cell analysis etc. Subtopics emphasized in this research include shape representation, measurement, and classification. A unified approach composed of contour based algorithms for solving problems of each types is developed, where contour techniques are potentially efficient because they involve relatively small quantities of data. Assuming that the input images have been properly binarized, contours defining distinct shapes can be extracted by following cracks (i.e., borders between black and white pixels). Shapes can then be aligned by matching their m-point contours. It is shown that given start point K on one contour, optimal scale, rotation and Euclidean distance, d(k), are non-linear functions of sample cross covariances at lag k between the x and y contour coordinate sequences. Contour models are data compressed shape representations, where the model parameters are contour transforms. An intriguing contour based shape recognition scheme, with classification cost virtually independent of the number of prototypes, is also presented. It utilizes complete shape information, is scale-translation-rotation invariant, and feasibly realizable.
- Research Organization:
- California Univ., Davis (USA)
- OSTI ID:
- 5606413
- Country of Publication:
- United States
- Language:
- English
Similar Records
EFFICIENT FEATURE-BASED CONTOUR EXTRACTION.
Implementation of a theory for inferring surface shape from contours