Home page for
Mark A. Robertson
Research: Image Dynamic Range
Suppose the day is nice and sunny, and for whatever reason you decide to take a picture from inside of a building looking outdoors. With your standard camera, you will almost certainly run into difficulties if you want to accurately capture both the scene indoors and out of doors. If your camera is smart, it might let you pick whether you want to capture the outdoors scene or the indoors scene; if you have a digital camera with a preview then maybe you can play with exposure settings to capture either indoors or outdoors areas. Odds are that in neither case will you end up with both the dark areas and the bright areas in a single picture. It is much more likely that you will end up with an image that looks something like one of the images in Figure 1 below, which were taken on a bright day in Cushing Hall at the University of Notre Dame. I don't know about you, but I wouldn't be happy with any one of these pictures!
Figure 1: From top left to bottom right are various exposures of a scene, with exposure time decreasing by factors of two.
The problem is that the dynamic range of your camera (to be more precise, the dynamic range of your camera's CCD elements or film) is not sufficient for accurately observing all the information in a scene such as that from Figure 1. The dynamic range of a scene range refers to the ratio of the brightest point in the scene to the darkest point in the scene. In a "normal" scene, dynamic range isn't usually a problem; for example, take a look at the image in Figure 2. Admittedly, there are some bright points, but there aren't any really dark points, and thus dynamic range is not much of a problem. Comparing the image in Figure 2 with those of Figure 1, it is obvious that the reason for the difficulty in Figure 1 is because it contains both bright and dark regions. The camera can only do good at getting just the bright areas, or just the dark areas; if you want both, you are just flat out of luck. Right?
Figure 2: Dynamic range isn't always a problem.
Well, that depends. One option for you is to just bracket your exposures---take a whole bunch of pictures at different exposures, and then pick the best of the bunch in the end. However, you are essentially picking among images like those of Figure 1. "Is there no other recourse?" you ask? Well, there are always other recourses, and they usually involve extra work on your part! Without getting into detail about what I have done (see my publications on dynamic range for those silly technical details), here is a short description of what to do: First take a number of pictures at different exposures (I vary only exposure time, to avoid difficulties with depth of field) so that all areas in which you are interested are captured in at least one of the pictures. The images in Figure 1 satisfy this requirement, although you wouldn't have to go to the bother taking quite so many. Then, line them up so that all image features are in the same locations in all images; this is called image registration. Some technical details must then be considered, but the basic idea is to merge all of the input images into a single image that has all of the detail from dark through bright regions. The merging must be performed carefully; for example, all of those bright regions (also called "saturated") should not be included in the merging, because they don't contain any useful information. Similarly, the dark regions do not contain much useful information either. Once you have a "high dynamic range image" you can play around with the contrast until you get something that looks good to you. Voilą! Below is a result from the scene in Figure 1. For the first picture, each pixel has been processed with the same transfer function, while for the second picture pixels in the doorway have been processed with a different transfer function than pixels elsewhere in the scene; thus the right picture has been processed in a "spatially-varying" manner. Admittedly, neither is exactly perfect-looking, but that's because, just like it is difficult to accurately capture a scene with high dynamic range, it is difficult to view a high dynamic range image on a medium with limited dynamic range (like, for example, your computer monitor or the printed page). But with a little manipulation of items like color balance, gamma, and brightness, it is quite possible to make the image look just the way you want it. There are more sophisticated ways of viewing high dynamic range images, but I have not implemented any of them.
Figure 3: High dynamic range results from input images in Figure 1, after some manipulation of contrast.
For the more technical details, check out my publications on dynamic range; my dissertation (in the publications page) also has some of my work on this problem. See the references cited in my publications for more sophisticated methods of viewing high dynamic range images. Also check out the following web locations for other peoples' methods of computing high dynamic range images:
Page last updated 7 February 2002