Precision Registration and Mosaicking of Multicamera Images

A method for registering and mosaicking multicamera images is described. Registration is achieved using control points and projective transformation, paying special attention to factors that contribute to the precision registration of images. Among the corresponding control points found in overlapping images, those that best satisfy the projective constraint are used to register the images. Experimental results show that a small number of correspondences that satisfy the projective constraint produce a more accurate registration than a large number of correspondences in least squares fashion. To achieve seamless mosaics, the intensities of all images captured by the cameras are transformed to the intensities of one of the images to minimize the intensity difference between the registered images. Mosaicking results on 4096 image sets acquired by a six-camera system are presented and discussed.