| 摘要 |
A system and method for improving the accuracy and convergence rate in determining the affine transformation of one image or array of values with respect to another image or array of values. A particular embodiment of the present invention comprises a reference and an input image. A gradient matrix, lambda, may be constructed to contain gradient information of the position values of the reference image. Also, an estimate matrix, rho, may be constructed to contain initial estimates of at least one of position, angle and scale of the input image. The input image is then subtracted from the reference image pixel by pixel producing an error matrix, e. The error matrix, e, is then multiplied with the matrices, lambda and rho. The result is the new change in the position, angle and scale of the input image. New values for the position, angle and scale are calculated from the changes in the position, angle and scale values of the input image. These new values are substituted for the previous values of the input image. The input image is consequently transformed. The transformed input image is then subtracted from the reference image pixel by pixel forming a new error matrix, e. A new estimate matrix, rho, is constructed from the previous estimate matrix, rho, the error matrix, e, and the gradient matrix. The process is repeated until, e, is less than a given value. When that occurs, the correspondence between all points in the input and reference image has been determined.
|
