Synthesizing a Panoramic Scene with a Common Exposure via the Simultaneous Registration of Images

An approach where a set of images can be simultaneously registered is presented such that these images are aligned both in their spatial coordinates as well as in their pixel values. Such an approach is capable of globally aligning a set of images to produce a panoramic scene with a common exposure and naturally addresses the misalignment problem common to pairwise registration approaches when image sequences loop back on themselves. The objective function considered minimizes the sum of the collective variance over pixels of a global grid on which to synthesize the scene. Our camera motion and lighting are constrained so that the images to register are spatially related by perspective transformations and so that the estimated exposure differences between images can be used to accurately synthesize an image with a common exposure. A detailed development of the mathematical foundations of the approach is included as well as results that demonstrate the method’s soundness.