| 摘要 |
Image processing herein reduces the computational complexity required to estimate a disparity map of a scene from a plurality of monoscopic images. Image processing includes calculating a disparity and associated matching cost for at least one pixel block in a reference image, and then predicting, based on this disparity and associated matching cost, a disparity and associated matching cost for a pixel block that neighbors the at least one pixel block. Image processing continues with calculating a tentative disparity and associated matching cost for the neighboring pixel block, by searching for a corresponding pixel block in a different monoscopic image over a reduced range of candidate pixel blocks focused around the disparity predicted. Searching over a reduced range avoids significant computational complexity. Image processing concludes with determining the disparity for the neighboring pixel block based on comparing the matching costs associated with the tentative disparity and the disparity predicted.
|
