| 摘要 |
GEODESIC SUPERPIXEL SEGMENTATION 5 An aspect of the present invention provides a method of segmenting an input image into a plurality of segments. The method comprises the steps of: deriving an image representative of boundary strength of each of a plurality of pixels in the input image (210); adding a random noise pattern to at least a portion of the derived image (230); determining a plurality of local minima in the derived image with the random noise pattern added (240), 0 each of the plurality of local minima comprising a point with a lowest measure of boundary strength within a pre-defined region in the derived image; and associating each of the plurality of pixels in the input image with one of the determined local minima to segment the image based on a geodesic distance transform of the measure between the determined local minima and the pixels (250). 5846288_1 -3/5 Start Derive boundary strength image 220 IAdd a fixed offset 230 Add random noise offset Xv' 240 Detect local minima as seed points 250 Compute geodesic 270 distance transform Reinitialise seed points to superpixel centroids2 Yes Frther efinement? No 280 End Fig. 2 5724332_1 |
