2D image shift registration through correlation and tailored robust regression

    公开(公告)号:US12125166B2

    公开(公告)日:2024-10-22

    申请号:US17864136

    申请日:2022-07-13

    CPC classification number: G06T3/4038 G06F17/11 G06T3/14

    Abstract: The disclosed method enables efficient registration and correction of random vertical and horizontal offsets of a plurality of images of a feature to enable combination thereof with maximum noise suppression. The method includes obtaining image cross correlations and estimating correlation peaks. If a correlation peak cannot be uniquely determined for a given correlation, all candidates are retained. An improved robust weighted regression is applied thereto to obtain estimates of the image shifts. Dimensional symmetry is exploited to remove redundancy from the weighted normal equation, and computational requirements are further reduced by analytically determining the coefficients matrix and the inhomogeneous matrix, thereby circumventing conventional computations. Solution of the resulting consistent, full rank, linear system requires only matrix inversion. For N images, the present method thereby increases computation speed and reduces storage by approximately N2.

    2D IMAGE SHIFT REGISTRATION THROUGH CORRELATION AND TAILORED ROBUST REGRESSION

    公开(公告)号:US20240020789A1

    公开(公告)日:2024-01-18

    申请号:US17864136

    申请日:2022-07-13

    CPC classification number: G06T3/4038 G06T3/0068 G06F17/11

    Abstract: The disclosed method enables efficient registration and correction of random vertical and horizontal offsets of a plurality of images of a feature to enable combination thereof with maximum noise suppression. The method includes obtaining image cross correlations and estimating correlation peaks. If a correlation peak cannot be uniquely determined for a given correlation, all candidates are retained. An improved robust weighted regression is applied thereto to obtain estimates of the image shifts. Dimensional symmetry is exploited to remove redundancy from the weighted normal equation, and computational requirements are further reduced by analytically determining the coefficients matrix and the inhomogeneous matrix, thereby circumventing conventional computations. Solution of the resulting consistent, full rank, linear system requires only matrix inversion. For N images, the present method thereby increases computation speed and reduces storage by approximately N2.

Patent Agency Ranking