Out of Core and Adaptive Image Blending Approach for Large Scale Image Mosaics

When creating large stitched images, like those used in maps made from aerial photos, it’s important to make sure the seams between individual pictures aren’t visible. This process, known as color blending, helps smooth out differences in lighting or weather across the images. But blending very large images, such as those made from many high-resolution aerial photos, can require huge amounts of memory, making it hard to do on a typical computer. In this work, we developed a method that breaks the problem into smaller pieces, so only a small part of the image needs to be worked on at a time. This makes it possible to blend very large images without needing an expensive supercomputer. Our approach uses a low-resolution version of the full image to help guide the blending of each small part, ensuring the final result looks smooth and natural. We also designed new techniques to make the process faster and more efficient, depending on how much blending is needed in each region of the image.