pic: Game hint

[Joe Ross]

Originally Posted by artdutra04

Eww, MS Paint. Those are some lonely megabytes on my hard drive, as I've haven't used MS Paint in about two years. The real story behind this image goes even deeper...

As an avid fan of Numb3rs, I like to dig deeper into everyday objects to unravel our most perplexing mysteries. I first opened the photo in Photoshop CS2, and cropped out the parts I didn't want (Otherwise the filter algorithms might return false positives). Then to highlight the differences in the photo, I ran the photo through a heightened saturation filter, with an offset in the hue. From this, I upped the contrast by about 20%, which now provided an optimal base image to begin the preliminary filter algorithms.

The first thing I would need to do was an inverse dithering and noise cancellation algorithm. Since the base photo was a JPEG image, there was noise in the photo from compressing it into a lossy file format; yet within this noise, there was also the image of the banana. Since the compression artifacts only effect the pixels around the source pixel, running the photo through these two algorithms simultaneously and integrating the results returned what I was looking for. I began to see the faint outline of a banana.

However, it took several tries to get a good filter pass of the photo. Because these filters act upon the image using Erwin Schrödinger's theories of exponential decay of pixels in compressed photos, sometimes the banana is still alive in the output, others it's not. And there is no way you can predict or know which is which, except for running the algorithm a lot and relying on an interpretation and statistical analysis of the results.

Using these results, I now needed another algorithm to be able to exactly detect the banana in the image, rather than rely on only human pattern-recognition skills, which in the hands of a creative person can result in one million different things visible in the same cloud in the sky. Obviously such a diverse range of output values given the extremely limited domain of input values in such a case would be extremely detrimental in our circumstance.

After doing some research into pattern recognition algorithms, and more specifically banana-recognition algorithms, I discovered this site about banana and apple recognition. It was a gold mine of information, and even had working algorithms. Bingo. My hunch about the banana on the left side of the image paid off. The algorithm predicted that is was indeed a banana with 99.7% certainty, and with a very narrow Gaussian distribution of the output data.

From there, I duplicated the layer and highlighted the banana in yellow to make it easier to see.

But after looking back through my source website again, so much of the website seemed applicable to FIRST at the moment. So with that in mind,


...I've thus concluded that this year's game involves... apples and bananas! How 'bout them apples?

I had to compress the photo more then default to get under CD's 750KB limit. I'm pretty sure it was a pineapple before compression.