Application of linear-scale differential analysis in phase correlation method of image stitching

A robust method of creation of panoramic images which does not consume much processing resources is proposed and investigated. The phase correlation (PC) method is taken as a basic one because of simplicity of its algorithm and low computing time due to application of FFT technique. Standard PC procedure is modified by preprocessing of source frames of panoramic images in spatial domain. Preprocessing comprises Linear-Scale Differential Analysis (LSDA) with sequent content-dependent thresholding of intensity gradients. Method is proved for artificially blurred and noise corrupted images. It is shown that new robust algorithm allows to increase the productivity of creation of panoramic images keeping the probability of successive stitching close to maximum even for low-quality source frames.