| 摘要 |
An efficient technique is disclosed for determining a portion of a document corresponding to a captured image. When a user employs a pen to create a stroke in a document, images of the document are captured by a camera mounted on the pen. While the location of some of the images will be determined from, for example, an analysis of a pattern on the document that is captured by the image or a pixel-by-pixel comparison of the image with the document, the location of other images will be determined by segmenting the sequence of images into groups that correspond to the shape of the stroke. Information relating to located images in a segment can then be employed to determine the position of unlocated images in the segment. For example, a document search region for an unlocated image can be established based upon the position of a previous located image and a maximum velocity or acceleration of the pen. The rotation and scale of the unlocated image are estimated as the same of the located image, and the unlocated image is warped using the rotation and scale. A pixel-by-pixel comparison can then be made between the warped unlocated image and the document search region. Further, if the warped unlocated image is matched successfully, the transform parameters of the image can be further refined. |
