Images (c) Dilip Krishnan & Rob Fergus

I was reading through the proceedings of this year’s SIGGRAPH (Special Interest Group on GRAPHics and Interactive Techniques) conference - as I’m sure we have all done at some point in our lives - when I came across a very interesting paper.

The conference generally concerns computer graphics and image analysis however the following title has a very practical application in photography: Dark Flash Photography. No we’re not talking about Joey Lawrence’s promotional images for the film Twilight, but instead a way to take flash photographs without people objecting, or even noticing!

The gist is that you can ‘hide’ the pulse of light outside the visible spectrum thus making it appear as much as 200 times dimmer than a regular flash. This requires an IR enabled camera and suitably modified flashgun. The use of flash for photography can range from an mild annoyance to a dangerous distraction in certain circumstances, which leads to the use of shutter speeds far slower than would be ideal. The dark flash solves this problem but creates a new one.

left to right, IR image, natural light image, combined image, long exposure natural light (for reference)

Capturing light in the IR part of the spectrum doesn’t allow natural looking colours to be recorded. Dilip Krishnan and Rob Fergus solved this problem by taking two exposures, one with the dark flash and one high ISO natural light photo immediately after it. The clever part is that the dark flash image can be used to guide to de-noise the high ISO image far more effectively that any other noise reduction technique. Simply put if you see an edge or spike in the natural light image and there isn’t a corresponding edge in the IR image then the point is probably the result of noise. The actual process is a little more involved, I’d recommend anyone technically minded to have a look at the original paper.

The result is a clean image with natural looking colours. The post processing required is complex, and there times when the approach wont work well if the subject moves between the two exposures (shooting fast paced sports for example). However if you have an IR enabled camera and like black and white images I can see this being a very valuable technique when in many circumstances, and indeed the only real solution to the problem of dazzling people with flash.

MG

MonaTweeta II
Originally uploaded by Quasimondo

Came across this interesting project today - the goal is to devise a scheme to compress an image so that it can be transmitted in a single tweet, which comprises a mere 140 characters - barely long enough to describe the four course meal you’re part way through consuming.

To illustrate the magnitude of the task at hand, that opening paragraph weighs in at about two tweets. Quasimondo’s solution utilises some neat tricks.

Firstly using Chinese characters increases the available storage as the set is larger than the standard western character set (which presumably enables anyone fluent in Cantonese to exchange essays over twitter).

The clever part as far as I’m concerned is the decoded image is generated using the Voronoi tessellation, which means each of the multi-sided shapes that make up the image can stored as a single point with x and y coordinates (plus a colour).

This enables large and potentially complex regions to be transmitted very efficiently. This does, however require the careful placing of points to generate salient shapes, whereupon the next clever idea is utilised. The points, or ’seeds’ are generated randomly at first and then mutated over many generations of a genetic algorithm to find the arrangement that best replicates the original image.

Click the thumbnail for a more complete description from the man himself.

MG