Abstract

Pyramids offer a representation in terms of a set of image primitives of many scales. Insofar as these primitives are appropriate to the basic tasks of image manipulation, pyramids offer a natural format for many operations. For example, in pyramid blending, parts of different images can be combined smoothly, without the artifacts associated with simple mosaic techniques. Smooth transitions are achieved at all scales, with a transition zone that is proportionately smaller as the scale of the image information becomes finer. In another application, multifocusing, one can merge a number of images focused at different depth planes (e.g., from a microscope), and derive a single image that is sharply focused at all depths. This is achieved by selecting the information of maximum local contrast at each position in each image, at each scale. Pyramids can also be used for impletion, the filling in of missing image information. Impletion occurs in natural images in blind spots and also in situations when only sparse information is available. Pyramid based impletion may be useful for image understanding tasks as well as image synthesis tasks.

© 1985 Optical Society of America