TRIX WITH PIX

Dick Pountain/Idealog 246/06 January 2015 14:05

My major digital pastime has for several years now been photography rather than programming: reading my profile reminds me I joined Flickr eight years ago and have now posted 1500+ pictures there (www.flickr.com/photos/dick_pountain/). The digital imaging market has been through a technical revolution during those years, and now faces what tech gurus love to call "disruption" thanks to the mobile phone. A whole generation now prefers their mobile to a proper camera, and phones' performance has improved extraordinarily by incorporating sensors and image processors from real camera manufacturers like Sony. Camera makers are striking back with gorgeous-looking retro designs that recall the golden age of the Leica, fitted with huge sensors, fixed "prime" lens and astounding image quality - and premium £1000+ prices aimed at separating "real photographers" from selfie-snappers.

As for me I've resisted both these trends. I started out posting mostly travel pics, street photos and landscapes, over-sharpening and saturation-boosting them to match the approved Flickr aesthetic, but in recent years I've become more and more interested in post-processing photos to make them more like paintings (abstract or otherwise). There are plenty of software tools available nowadays to spice up photos - some like Google's Nik Collection of plug-in filters for Photoshop and Light Room are very good indeed - but I'm less interested in buffing up my pics than in dismantling and reconstructing them completely. And my chosen tool is therefore, er, Photoshop Elements version 5. This ancient version lacks all the smart cut-out and similar features of later versions, and many abilities of full Photoshop, but it has all I want which is basically layers, blend modes and a handful of filters.

My modus operandi is as eccentric as my choice of platform. I perform long sequences of operations on each picture, duplicating and saving layers, tinting, filtering and blending them in different modes, but rather than write down this sequence so I can repeat it I deliberately do *not* do that. I merely watch the continually changing image until I like it well enough to stop. I can never repeat exactly that effect again, which I've convinced myself makes it "art" rather than mere processing, just as an oil painting can never be exactly repeated. Doing this so many times has given me a fairly deep grasp of how pictures are made up, about manipulating different levels of detail and tonality. One of my favourite filters is High Pass, which can separate out different levels of detail so that you can enhance or remove just that level. Another favourite trick is mixing some percentage of an outrageously processed image back into the original to temper the effect and make it more subtle.

In view of all my coal-face experience of the internal makeup of digital pictures, I was interested to hear about a joint project by GCHQ and the National Crime Agency (NCA), announced in December 2014 by PM Cameron, to deploy new recognition algorithms for identifying online pictures of child abuse, to aid in their prosecution. The press release said these algorithms are "hash based": that is, they process the bitstream of a digital picture to reduce it to a single number that becomes a "fingerprint" of that picture. Such fingerprinting is essential for evidence to be acceptable legally: it's necessary to prove that a picture confiscated from some offender is the same as one obtained from someone else, and obviously filenames are of no use as they're only loosely attached properties that can be easily changed.

The US website Federal Evidence Review suggests an algorithm called SHA-1 (Secure Hash Algorithm version 1) is in use for this purpose, but it appears to me that algorithm is designed for use on texts, gun serial numbers and other alphanumeric data sets, and I can hardly believe it would generate useable hashes from bitmapped images whose contrast, saturation, sharpness and so on may have been altered - either deliberately during enhancement, or merely by accident through repeated sloppy copying of JPEGs. Pictures that are perceptually similar might have bitstreams quite different enough to change the hash.

I'd guess that content analysis, not merely hashing the bits, will be needed to prove the identity of two versions of any bitmapped image. Face recognition is well advanced nowadays (recent compact cameras can even distinguish smiles) and so is dissection of bitmaps into separate objects in Photoshop. It would remain challenging to create a unique hash from the collection of persons, furniture and stuff isolated from each picture, and oddly enough it's in fine art rather than criminology that the required expertise is most advanced. Iconclass is a hierarchical notation developed by Dutch painting scholars for cataloguing unique configurations of picture elements, and what's needed is something similar for far less salubrious subject matter.