| 摘要 |
A system and method for representing and encoding an input image. Pixels associated with the structural part of the image are referred to as "sketchable," and pixels associated with the texture part are referred to as "non-sketchable." A sketch graph, Ssk, is generated for the sketchable part of the input image by iteratively selecting a plurality of optimal image primitives from a dictionary of image primitives. The sketch graph may then be refined to achieve good Gestalt organization and resolve ambiguities in the graph. The non-sketchable texture regions of the image, Snsk, are computed via a simple texture clustering algorithm. In reconstructing the sketchable part of the input image, the image primitives are aligned on the sketch graph. In reconstructing the non-sketchable part of the input image, the MRF models are simulated to reproduce statistical summaries of the various texture regions using the sketchable part as boundary condition.
|
