Morphological scale-space analysis and feature extraction

This paper presents a morphological scale-space approach to the problem of feature extraction. The method relies on two steps: a hierarchical simplification step based on pyramids of morphological operators and a feature extraction step consisting in measuring the persistence of each image structure through the simplification scales. Specific scalespace properties are needed: the features should be ranked in a monotonic way and the contours should not be corrupt. Adequate scale-space operators are designed according to these properties. Depending on the filtering criteria on which they are build, a variety of attributes of the objects in the images may be extracted: the size, the shape, the contrast... Different examples illustrate the usefulness of this strategy.