Your cart is currently empty!
How to See Like a Machine
We are in a brave new world of photography; we can leverage machine learning and artificial intelligence to make better photos! Some things I’ve been learning:
DOWNLOAD
PDF:
Also download photo examples:
Behind the Scenes Video:
The making of video, with screen capture and Camtasia on my laptop:
Image Segmentation
To better study and look at images, segment the images (something I learned from machine learning):
How do machines see images?
To start off, I want to state that humans are far superior to machines in almost every way. Let us not get fooled by the ‘golden calf’ of AI (artificial intelligence):
The purpose is for us to leverage machines to make better humans.
Not,
Trying to use humans to make better machines.
To re-iterate:
Let us leverage machine learning to better improve ourselves.
So I don’t think the purpose is for humans to see like machines (remember Arnold S as the Terminator, when he sees the world in black-and-red). What I propose is that we study how machines see images, as a way to guide how we can better see as humans.
Or in other words:
Let us study machine learning, to better understand how we (as humans) think and see!
Simplify
The interesting thing about how machines see images:
They simplify it.
When I study my own black and white photos, I believe my best photos are my simple photos. Even if you study great artists like Picasso and Matisse, they always strived for elegant images– simple and elegant. Think of Matisse’s cutouts — he was able to abstract the human body into the simplest shape, form, and color.
Figure-to-ground
Figure (the subject) and ground (background): you want maximal contrast between the two elements.
One thing we can use: “Gaussian Blur” in Photoshop, to abstract the images, to just see the position of our visual elements, and to simplify the contrast:
Does the photo work well as a small thumbnail?
To better understand whether your compositions are strong or not– look at them as small thumbnails. If your photos work as small thumbnails, it probably means your composition is strong!
Bounding Boxes
There is already artificial intelligence that can automatically put ‘bounding boxes’ (little red boxes) around subjects. This is how self-driving cars see pedestrians (not to run them over):
Even more insane– machines can automatically color, segment, and classify different subjects in a scene (real time):
A simple way we (humans) can look at photographs:
Let us put little red boxes around different visual elements in photos.
The Cherry on Top
For example, I like to look at the ‘cherry on top’ for subjects in the photograph– small details which I find interesting. For example, in this photograph, this little boy in the background is a detail I love:
But this is the difficult thing– most people (when they see the picture really small, perhaps on a phone), they won’t see the detail of the boy in the background:
This is why I like printing out photos big– so you can actually see small details!
And to be honest, 99% of your viewers won’t notice these ‘cherry on top’ details in the photo. But it doesn’t matter– as long as YOU see the detail and like it– that is enough!
Color Segmentation
We can also apply gaussian blur to color photos, to better understand the color palette of our photos:
I used the color picker in Photoshop (hotkey “I” for “I[eye] dropper”). And painted these colors:
If I make the background grey, you can see the colors:
So you can see the basic colors: rogue red, pale yellow, and green in the bottom-left.
Keep the colors simple
For color photography, I generally think the simpler, the better.
For example this picture of this woman, the primary color is pink (with Gaussian Blur applied):
The final with the background subtracted:
Which details do you look at in a photo?
If you use the concept of a ‘bounding box’ — you can use it as a tool to identify things you find interesting in a scene– like the woman smoking a cigarette, and two Red Bulls in her left hand.
Feature Map
When you are shooting photos, try to abstract your vision — simplify it.
For example, this is why I recommend shooting high-contrast black and white mode, because it helps you see better when you’re taking photos.
For example on a RICOH GR II, I recommend using the ‘high contrast black and white mode’ in JPEG, and when you’re composing your photos, keep the photos simple.
Silhouette
This is how silhouettes work — we create a ‘feature map’ (the gradients between black, grey, and white.
Edges
Also another way to look at photos; try to simplify the edges, to see the textures in the photo:
Silhouette of Faces
Another idea:
As humans, we intuitively know the proportions of a human face, and also the silhouette of faces.
To see better, let us apply maximal contrast to the photo:
Then, a cyan cutout of the woman:
Conclusion: The Future of Photography
I’m currently doing more studying on Augmented Reality (AR), especially the ‘AR CORE’ platform from Google, as well as their “Tensor Flow” platform. The basic idea I have is this:
One day, while you’re taking photos on your photos, you will get compositional grids, and ‘bounding boxes’ to help assist your compositions.
I already have this functionality on “intelligent Auto” mode on my Lumix G9– when I am composing a scene, the LCD screen will automatically put a yellow box around my subject in the frame.
Augmented Reality for Photographers
I think the future of photography is this:
Photographers will leverage artificial intelligence (AI) and machine learning to take better photos (improved composition, or awareness of subjects).
The future is already here
For further reading, I recommend:
- Research “AR CORE” from Google
- Research ‘computational photography’
- Check out the Google AI Experiments
- Download some “AR” (augmented reality) apps on your phone or iPad and play around with it
- Think about how photographers, individuals, and artists can leverage virtual reality, augmented reality, and machine learning to make better art!
ERIC