A Study upon Sub-Pixel Accuracy of Point Identification between Digital Images

An idealized model on re-constructing digital image between coarse and fine pixel image is proposed. The maximum local cross-correction coefficient method is employed to identify the associated point of an digital image with respect to another digital image. The distance between two corresponding points is considered as the displacement between points. After comparing this distance to the exact distance, the identification error can be defined. It is found that the following methods can improve the error on identification: scale up the gray level value by an amount approximately equal to the maximum gray level of two images; construct the fine pixel image via the monotonic cubic spline interpolation to perform the fine grid identification; and employs a modified Shepard interpolation to perform selected interpolation which excludes all points with an extra-ordinary displacement. The linear conservative interpolation method is examined and is found to have a small effect of improving identification error because the proposed method of calculating the gray level gradient is improper. Ocassionally, it is helpful to improve image visibility. A motor plate image was employed to test the applicability of the proposed model. The result shows that only the gray level scale up and the selected Shepard interpolation are effective.