Gradient Domain Image Blending and Implementation on Mobile Devices

This paper presents an image blending approach which combines optimal seam finding and transition smoothing for merging a set of aligned source images into a composite panoramic image seamlessly. In this approach, graph cut optimization is used for finding optimal seams in overlapping areas of the source images to create a composite image. If the seams in the composite image are still visible, a gradient domain transition smoothing operation is used to reduce color differences between the source images to make them invisible. In the transition smoothing operation, a new gradient vector field is created using the gradients of source images and the seam information. A new composite image can be recovered from the new gradient vector field by solving a Poisson equation with boundary conditions. Our approach presents several advantages. The use of graph cut optimization over the source images guarantees that optimal seams are found. The gradient domain transition smoothing operation allows smoothing out color differences globally and further improves image quality after merging with graph cut optimization. The final composite image is a global optimal solution. The approach is implemented in two ways called sequential image blending and global image blending. Sequential image blending allows us to use little memory in the whole blending process, which is very important for mobile devices. Global image blending guarantees a globally optimal solution. Experimental and application results in creating mobile image mosaics and mobile panorama are also given.