| 摘要 |
The present invention relates to the use of an electronic camera in a platenless document imaging system in which the document image is a composite image formed from a mosaic of overlapping images (51,52) captured by the camera. Residual misalignment between overlapping mosaiced images is reduced by identifying in at least one overlap area (50) along adjacent edges (68,70) of adjacent image tiles (51,52) the locations of pixels representing one or more matched pairs of identifiable image features (75,95) present in both adjacent image tiles (51,52). Error vectors (98) are generated associated with each matched pair of features (75,95), a matched feature error vector (98) having the property that if the location in the current image tile (52) of a pixel (85) associated with an image feature (75) is shifted according to said error vector (98), then the locations of the same image features (175) in the warped current image tile (152) and warped adjacent image tile (151) are brought together to reduce the misalignment. A full set of error vectors (100) is generated by extrapolating and interpolating from the matched feature error vectors (98), extending along a corresponding edge (68) of the current image tile (52) in a direction transverse to this edge (68) towards an opposite edge (63), with each error vector (100) in the full set of error vectors having a magnitude with a contribution from the matched feature error vectors (98) that is progressively reduced away from said overlapping edge (68). The error vectors (100) are used in an adjusted transform based on an initial transform to reduce misalignment in a composite image of the document. <IMAGE> |
