**Testing Stereogram Interference Limits and Depth Perception**

**Philosophical Quandary of a Deep Neural Network**

I'll be the first to admit that Stereograms have been around for a long time now and they don't exactly constitute art. However. I have been thinking about Deep Learning and how computer software will be reading and categorizing images in different ways than humans -- I started making some stereograms on Photoshop because I was interested firstly, in understanding how retinal disparity works, but also in steganography. I was wondering first of all, if I made a block color image of Blue but laced a Red block of color into the ASCII code, would image recognition software also detect and tag that block of color as Red, while the person looking at it clearly only sees Blue?

I first made a stereogram with a repeating binary text conversion, but I wasn't getting the same image results as when using a random dot stereogram generator. Then, I made another one with some ASCII text trying to figure out in which ways I needed to repeat the pattern to achieve the best depth perception of the hidden image. In the image above, I just hid a circle in the text pattern however when I look at it, I see an undercut as well as an overcut of a crescent shape -- so, I'm still working out how the patterns need to repeat so that I can represent the hidden shape properly.

Following on from that, I discovered a few interesting things (although in hindsight, are maybe obvious). Random dot stereograms generated with a number of colors can be reduced into one color only and still display the hidden image. I also found that the stereogram can still be visible with quite a bit of interference blocking out the displaced pattern. I discovered that I could increase the 'pop out' effect of my overlapping sketch when placed over the hidden image -- I suspect this has something to do with the high contrast but am still working out the nuts and bolts of it all.

Finally, on a philosophical note, I really was just left wondering. In the end, does the Deep Neural Network really 'see' and compute more than we can, or is it the human that really perceives more?

Interesting questions, particularly for an artist to grapple with, I think.