| 摘要 |
A method provides a compact and robust representation of the content of images or video frames, for use in image retrieval, recognition and classification. The method comprises processing signals corresponding to an image to identify a plurality of feature points in the image, and deriving feature descriptors of feature points. Feature descriptors are assigned to pre-defined centre points, wherein each feature descriptor (ldn) is assigned to a plurality of centre points (Ck), thus increasing the number of local vectors assigned to each centre. The method further comprises, for each centre point, calculating the difference between each feature descriptor assigned to said centre point, and deriving a value descriptor for each centre point from said calculated differences. The representation is derived from said value descriptors for said centre points. Deriving a value descriptor for each centre point from said calculated distances may comprise transforming each distance by a robust function. |
