High precision sub-pixel spatial alignment of digital images

摘要

A high precision sub-pixel spatial alignment of digital images, one from a reference video signal and another from a corresponding test video signal, uses an iterative process and incorporates spatial resampling along with basic correlation and estimation of fractional pixel shift. The corresponding images from the reference and test video signals are captured and a test block is overlaid on them at the same locations to include texture from the images. FFTs are performed within the test block in each image, and the FFTs are cross-correlated to develop a peak value representing a shift position between the images. A curve is fitted to the peak and neighboring values to find the nearest integer pixel shift position. The test block is shifted in the test image by the integer pixel shift position, and the FFT in the test image is repeated and correlated with the FFT from the reference image. The curve fitting is repeated to obtain a fractional pixel shift position value that is combined with the integer pixel shift value to update the test block position again in the test image. The steps are repeated until an end condition is achieved, at which point the value of the pixel shift position for the test block in the test image relative to the reference image is used to align the two images with high precision sub-pixel accuracy.