Abstract:

A highly robust stitching algorithm of panorama is proposed to improve the efficiency and accuracy of building panorama. Firstly, a pyramid-based image is formed accoding to the input images, and then the least local entropy and a sequential similarity detection alogrithm with an adaptive threshold are used to complete the stitching of the panoramic images. The experiments show that the method has the properties of anti-noise and anti-geometrical deformation. And it indicates a high robustness to the images taken by a camera in common conditions.

Key words: virtual reality, local entropy difference, sequential similarity detection, panorama