| 摘要 |
[Problem] To allow carrying out, automatically and with high precision, region segmenting of an image with a graph cut. [Solution] A candidate region of a region to be cut is set upon an image. In a graph comprising: a point (S) corresponding to a region to be cut; a point (T) corresponding to a background region; points corresponding to each pixel of an image; S links which join each pixel to the point (S); T links which join each pixel to the point (T); and N links which join adjacent pixels; costs of respective links are set; and a graph cut executed. Thereafter, a determination is made as to whether pixels are present in the graph which are joined to the point (S) by the links; and if they are not present, the graph cut is executed in each level while incrementing, by levels at intervals less than or equal to a predetermined threshold, the costs which are set to all of the S links which join each pixel to the point (S) in the candidate region, until the pixels appear which are joined to the point (S) by the links. According to the results of the last executed graph cut, a determination is made as to whether each pixel of the image is part of the region to be cut or the background image. |
