The first steps for this project were to capture the images, define correspondences, and recover homographies.
I used my calculated homographies and the method of inverse warping to 'rectify' images. I essentially warped the image to a shape that would change the perspective to appear as if we were looking straight on.
Here I aligned the warped image to the unwarped image. From there I blended the images using a weighted average. For each pixel I calculated its distance from the center of that image, weighted the value of the pixel from each image by 1/distance, summed the values and divided by the total of the two scalars. This gives greater weight to those pixels closer to an image's center. As we can see the results do not look good, and this is due mostly to a misalingment. Additionally, better methods of blending could be explored, such as through the use of Laplacian Pyramids.
There is still much work to be done to improve the results. Additionally, the difficulty of this project makes me appreciate the simple panoramic settings on cameras.