Assignment 1

This assignment was to combine several images of the same scene, captured into varying levels of exposure, into one image with all the details visible, using a process called tone mapping.

I preformed this on two sets of images, one which I captured myself, one which was provided. First, let me present the provided image of an igloo-ish thing (what exactly is this thing? I have no idea.).

As you can see, in the darker images, the background details are all visibile, but we cannot see inside the igloo anymore. To produce a combined image, we first average all six of the images above.

Now, lets look at producing an HDR imaged. This can be done with HDRshop \ and this is the resulting image. Then, we produce a tone mapped image, to get the results with eight bits per color channel. Below we have the intermediate base and detail results, followed by the final tone map.

Not quite perfect, but atleast we can see both the background and what is inside the igloo.

Now, the sames steps I apply for the images I captured myself. This is a picture of some construction going on outside my dining room window. Tufts is doing extending a parking lot, apparently. Now I will live slightly closer to campus. You can see some of their buildings in the background. There is a set of candles on the table on a shinky metal plater, so there is quite a range intensities. Click on any of the images to see the full size version.

Now, the average image.

The average is a little bit blurry. Why? Because I couldn't get my camera to remain still while I adjust the exposure-- my tripod was designed to hold those ridiculous directional 802.11 antennas, not cameras. I'm no photographer. This (56 MB) is the HDR image produced by combining the above. Next, we have the base, detail, and final tone mapped image:

Now we can clearly see the candles as well as the smoke stack in the background! But it still looks a little dark. I can make it brighter by tweaking the parameters, but then the candles turn red. I'm not sure how to resolve this. pErhaps a better tone mapping algorithm, or maybe taking better pictures in the first place? Well, finally, here is the matlab code I wrote that does this:

img_avg.m. This takes a list of file names, and returns the average image in RGB, with pixels in the range [0,1]. The images had better all be of the same size!

tone_map.m This does the tone mapping. The filter argument is the smoothing filter to apply (I used a gaussian, with default parameters in the first images, and with 50,16 second), and the contrast is the amount to reduce the contrast (valid values are [0,1]). I used .1 .05 to produce the results above, respectivly.