What is HDR Deghosting?

HDR (or HDRI) is short for "high dynamic range image". Images we see on our computer screens range in brightness from 0 to 255, but there are many more intensities of light in the real world. This is why when we take photographs of magnificent scenes, like sunsets, it often turns out dull. High dynamic range images contain a higher range of light. The values HDR images can be processed by tone mapping to be viewed on regular monitors.

To capture an HDR image with a regular camera, one can take several pictures of the same scene, adjusting the shutter speed with each exposure, resulting in a series of pictures from dark to light. An HDR image can be made by combining the well-exposed pixels from the source images. However, there are frequently moving objects (such as people) in the scene, causing each picture to be inconsistent with the rest. This results in ghosting - objects appear semi-transparent. The process of removing the semi-transparency is called deghosting.

More Khan Results (the bad one)

There are some cases where Khan's algorithm doesn't work at all. In the previous post (the fountain scene), the tree branches could not be deghosted because the wind movement caused each image to be different, preventing the algorithm from finding a background. Even with large movements, as long as the images are different enough, the algorithm will have trouble. One example of such a case is when taking pictures of crowded places.

Source images:

Even though this scene is far from crowded, there's a lot of overlap in where the movements occur.

After 1 iteration:

After 5 iterations:

After 12 iterations:

The only difference between the first iteration and the 12th iteration is.... the error was more apparent in the latter one =P. One way to attempt at a solution is to encourage choosing from the same image when the weights are indecisive, but then the order in which the pixels are processed becomes very important, and when will we know it's ok to switch to a different image? We can't simply choose the input with the largest weight, because that could cut into some objects. For example, the body of the man on the right has a relatively high weight, while his head has a lower weight than the wall, and we don't want to decapitate him. This will require calculating what's the most coherent "edge" for switching images, which is very slow. We could use the calculated weights to help make the decision (similar weights = better seams), but I haven't sorted out the details yet...

No comments: