why the real photo quality of an image cannot be (yet) calculated by a machine
How do you identify the photo quality of an image? Is there a mathematical algorithm that can tell us which are the best photographs amongst a collection?
I want to share my experience as a photographer and my doubts as to the ability of super-duper artificial intelligence to identify the true quality in a photo. Images are personal, and need a personal view to identify their true… value.
This is one of the elements we have worked on while bringing phlow to life. We wanted to foster compelling images, images with a photo quality that was not related to pixels, exposure or any other cold variable. We want images that could draw out emotions and resonate with viewers.
We wanted to organise streams of images by a photo quality that connected with your soul.
ask 10 photographer to select the photo quality of your portfolio and you will get 11 different answers
If you are a photographer, pick up your portfolio and show it to ten different professional photographers. If you are not a photographer, just pick a selection of images and ask the “pro’s” what they think of the images you selected…
What you will realise is that you will get a lot of different opinions, and I am not talking about subtle differences of view from different photographers; I am talking about completely contrasting views.
When I was starting my photography career, I asked few photographers to criticise my portfolio. I ended up more puzzled and with more questions than I had before the critiques. The photo that for one was the strongest, for another was the weakest. The beautiful details one of them saw were a distraction for another. The photo quality was never a constant.
”beauty is in the eye of the beholder”
To have one artificial intelligence identify the photo quality is like following the feedback of one photographer. What about the other nine? What about the subtle, emotional differences an AI can miss? What about the fact photo quality is personal to the viewer?
the emotional element of a photo quality
There are many lessons I have had to learn while growing my own photography business. One of these is that to be a good photographer, you need to be a good editor. To decide which images to show your clients is as important as the ability to capture those photos in the first place.
In the early days I focussed on picking a “technically good photo quality”; sharpness, composition, exposure and other technical mumbo jumbo were at the base of my selection. And do you know what? The day I decided to show an out of focus, artistic image, I sold it. It had no technical photo quality, yet the client paid to have it.
The emotional element of an image is one of the driving factors of photo quality. Yet, this part of a photo is not something that can be identified, as it is not an objective value. The photo quality then becomes a subjective variable, that revolves around the viewer more than the photo itself.
Henry Cartier-Bresson, one of the most renowned photographer of the twentieth century, focussed on “the moment”. Some of his iconic photographs captured something unique, that cannot be understood by a machine. So how then can we expect to find the true quality of an image without these subjective elements?
perfect composition and exposure VS real life
A few months ago I decided to try a service from eye’em known as “The Roll” that scores the images within the camera roll of my phone. I let it analyse everything I had in there, whether quick snaps, screenshots or some of what I considered my “iconic photographs”.
I was curious to see how an advanced code could have scored the photo quality of such different images. To be honest and I don’t want to do this clever app any in justice, I was impressed by the ability of it to identify my photographs. However, what I realised is that the scoring of the images had nothing to do with the true photo quality I was seeking. Some of my black and white silhouettes were compared to noise, while some really ugly snaps were shown as the best photos I ever took.
The reality is that there are limits to current artificial intelligence and how it perceives the world. Despite leaps in quality, real life is still far from being understood by them. A photo quality is still personal, emotional, and defined at a deeper level.
So how can we do it better with phlow?
AI VS Behavioural Analysis in understanding photo quality
Artificial intelligences have come a long way, as I mentioned above. They are much better at understanding the contents of photographs, yet they still fail at identifying the meaning of those contents.
Flickr and its “home page” feature drove photographers crazy for years; while nobody outside Flickr really knows for sure, the general consensus is that the algorithm was taking into consideration a behavioural analysis to calculate the photo relevance.
The issue with this, which has since been replicated verbatim by 500px, Instagram and many others, is that the behavioural analysis is limited by a friends circle. So photographers started playing “like4like” and “comment4comment” games, which ruined the logic. Instead of identifying photo quality, these services were rewarding photos published by users with more active friends’ circles.
This is why in phlow you don’t follow friends. You don’t define whose photos you want to see, you define the context of photos you want to see.
In my favourites I have #blackandwhite #portrait, #yoga and #surf, because those are the things I want to see. The more I “phlow” those streams of images, the more phlow can understand my behaviour and merge it with the behaviour of other readers who share similar passions.
We have created a virtual curator whose role it is to identify the photo quality for each image within each context based on crowd sourcing. It is not an AI, it is a system that listens to what you think about the images we have.
An image can be completely out of focus, yet it many have deeper meanings.
So here is why we haven’t implemented an artificial intelligence: because we don’t believe it can tell us much about how compelling photos are.
Can phlow help develop the future AI to help us identify photo quality?
The photo quality is something that, one day, will be understood by an artificial intelligence, I am sure about it. Technology will grow so much that, sooner or later, will be able to understand context and emotional meaning of images. Yet, we are still far from this.
My goal as photographer is to be seen for the quality of the photos I produce, not for my ability to network. My goal as founder of phlow is to evolve our system in order to start understanding more about photography than any other system.
I am less interested in the technical value of an image and more in the emotional elements of a photo quality. I believe that understanding the human response to an image in a system that does not put the social connections first, can lead to a breakthrough in this industry. This is, of course, a long term goal I have as entrepreneur. For the time being we are trying to create the best place to consume the images we love, images displayed on their true emotional photo quality, not the cold technical details.
