June 17: Is 50 megapixel enough? Just how much resolution is needed to make a picture?
Jean-Jacques Calbayrac has an appropriately small presentation, GAMEBOYCAMERAMAN, of unusually low-res photography on show in the lounge of the Hague Museum of Photography until September 10.
He takes photos using the Game Boy Camera (GBC) an official add-on cartridge for the handheld Game Boy gaming console developed by Game Freak Inc. (株式会社ゲームフリーク) who are the producers or Pokemon, and released on September 17, 1998 in Japan.
Before Nintendo ceased its manufacture in late 2002, their Game Boy Camera was included in the Guinness World Records book of 1999 as the world’s smallest digital camera. It takes 256×224 pixel images (down-scaled by the screen to half resolution, 190×144 pixels, with anti-aliasing) in black & white (grey/green on the screen) using the 4-colour palette of the Game Boy system.
Pictures made with it by Neil Young’s daughter Amber Young (*1984) featured on his album of 2000, Silver and Gold.
So there you go…it’s a professional camera!
Calbayrac found his Game Boy Camera in a Paris flea market ten years ago, and uses all of its 128 x 112 pixels (0.05734 megapixels!) to photograph his surroundings in London with the aim of making “the line between familiar and new…blurred.”
Now he has around a half dozen of the retro gaming peripherals, and he carries all of them when he’s working around London. He releases the images on his Instagram feed where they are distinctive for their coarse pixellation amongst the smooth hi-res phone pictures, like those from the latest iPhone camera’s more than 12 million pixels, posted by other users of the social media app.
When I was young I dreamt of having it, but I couldn’t afford it. Now I’m taking care of this little cube very, very carefully, because if it breaks I’m just going to be lost. I don’t know what I will do.
In the spirit of the makeshift quality of the images, he has been pasting enlarged prints on walls around Spitalfield’s Brick Lane as well as in Columbia Road and Rivington Street in in neighbouring Bethnal Green:
It’s not an exhibition – it’s in the street, by myself, with no champagne for the opening, but it’s not a problem.
The poster shows the classic gasometer, the larger one of a pair, beside the canal at Bethnal Green, built 1888-9 to a design by George Trewby.
The low resolution turns the people in the photographs into mere cartoonish ciphers that emerge from clusters of dots, crosses and patches of tone, but seen from a longer range the brain fills in the detail to give a quite nuanced tonal image. Unknown or unfamiliar objects (is that a newspaper vending machine next to the man above?) however do not resolve and remain mysterious, a result that accords with Calbayrac’s aims.
What humans can see in a Game Boy Camera image, machines would have greater difficulty identifying and it is a skill we have each acquired through learning. In simple tests of preference, infants as young as newborns prefer faces and face-like stimuli over other visual stimuli, and between 3 and 9 months of age, infants gradually focus their attention on faces.
Roland Meertens is a research engineer in a translation software in Neural Machine Translation that uses neural networks to estimate the quality of a translated sentence. He has applied convolutional neural networks – machine learning – to the task of producing photorealistic images from Game Boy photos.
Refining the images is of course quite counter to Calbayrac’s aims in adopting the camera, but it does throw light on the way human perception operates and it is of course relevant to the emerging field of face recognition in security applications.
Researchers Yağmur Güçlütürk, Umut Güçlü, Rob van Lier, and Marcel van Gerven have implemented neural networks for solutions in ‘super-resolution’ (upscaling an image without loss), appropriately colourising greyscale images, removing JPEG compression artifacts. Meerton was inspired in particular by the feat of turning sketches into ‘photographic’ face images, deciding to apply it to Game Boy camera images of faces and turn them into photorealistic images.
He achieved good results by ‘teaching’ the network with images photographed with the Game Boy Camera, and asking it to create a new image by from that and a full resolution image. Finally it was possible for the network to realistically colour the built-in Nintendo game image alerts; manipulated funny faces:
When applied to a ‘selfie’ photographed from the screen (he had no way of transferring the Game Boy Camera data to his computer), the result is a low-contrast but acceptable ‘selfie’.
Another Game Boy Camera user was able to obtain a transfer directly from his camera and sent it to Meertens who applied his neural network to work on it, with this final, quite convincing result in which a machine achieves what humans do with Jean-Jacques Calbayrac’s Homme au Hasard XVII above, with the added bonus of skin colour.
You can read the whole description of Meerten’s research project here.
Jean-Jacques Calbayrac (*1987, Menton, France) took up photography when inspired by a video presenting street photographer William Klein in the Contacts series. He started travelling to shoot pictures and in 2010 he joined the Gobelins in Paris from which he graduated in June 2012.
It’s not just a game!