How is machine learning transforming modern photography?
No longer just a sci-fi movie plot twist, artificial intelligence (AI) and machine learning (ML) impact our daily lives in many ways. Watch Netflix? When you get recommended shows and movies that’s AI. Work in marketing? Machine learning is behind the predictions an ad platform will make about how much money to spend to get a certain amount of user clicks.
Simply put, computers use AI to do things typically done by humans, or simulate human intelligence by machines. Computers gather loads of data and AI analyzes that data and acts on it. Machine learning is a subset of AI that uses the data to imitate the way that humans learn, gradually improving the AI, all on its own, without being programmed in a specific way. The machine learns from what it experiences.
Creativity, an innately human concept where we use our imaginations, is an unexpected area that has blossomed with AI and ML. “It helps to automate mundane tasks, allowing creatives to spend more time on inspiration and design,” said Chris Duffy, Adobe Creative Cloud strategic development manager in his AI Extends the Creative Mind article.
And perhaps the creative discipline transforming the most with AI and ML is photography, giving photographers of all levels the ability to create better photographs and compositions, quickly and efficiently.
Editing images
Capturing the perfect image is only the beginning — whether you’re a novice or a pro photographer — you need the right editing options to make your photos amazing, Editing features include changing unwanted backgrounds, applying unique looks to photos using presets, correcting colors, changing brightness levels or contrast to alter and enhance images.
There are so many editing options now that it can be daunting for photographers to know exactly how to make an image look just right. With AI and ML features built into photo editing software, a wide array of editing options become available, and easier to use. For example, you can brighten photos, bump up colors, and mask distracting elements quickly and efficiently. AI and ML helps you fully transform images with a few simple clicks, eliminating time-consuming tasks.
Below we dive deeper into key image editing features where AI and ML make dramatic differences to help photographers.
Image enhancement and definition
Image enhancement is the process that improves picture quality without information loss so that the results fit the creator’s desired resolution, color, and style. This involves several transformations to photos such as contrast enhancement, deblurring, removing visual distortions and so much more. Learning to execute these types of enhancements takes time, and the quality of the results depends on the skills and judgment of the editor, making it subjective. Using AI photo editors, creators get the expertise of a skilled retoucher — taking away the manual work — while providing results that are still under the editor’s control.
For example, photos shot in Raw format (minimally processed data) are stored as a set of instructions rather than actual images, and machine learning uses the color data for each pixel to guess what the final image should look like, dramatically improving the quality of the raw file.
Deep learning, a subset of machine learning, can be applied to image enhancement. Previously, taking an image from low to high-resolution involved going through the process of taking an entirely new photo or rescanning the existing image. But with enhancements like neural filters, which are powered by deep learning technology, it’s possible to create pixels from a small set of image data and combine them with the existing pixels to transform images from low to high-resolution.
Lighting and coloring
Artificial intelligence also plays an important role in the aspect of lighting and coloring. Making hand-colored images is slow work, but AI lets creators colorize their photos with a few clicks. With machine learning, photo editing apps are trained to identify both unique and predictable colors of different objects, and apply these colors to black and white photos.
Cropping and filling
When viewing images, many people tend to focus on the essential parts of the photos at first glance. With machine learning, neural networks can also be trained to detect these significant elements, making it easy for creators to identify and select what part of their images to crop, as well as filling in deleted parts of the photos.
For example, if you want to make a collage of several images, the first thing you have to do is select and cut out objects, and no one wants to waste time cropping hundreds of images into the right format. In Photoshop, the AI-enabled selection tools feature highlights the obvious objects in your image, thereby allowing you to move them seamlessly.
Machine-generated images
Machine learning generated images are photos created from generative adversarial networks (GANS), which are a set of algorithms that are trained with thousands of real pictures to produce artificial images that look realistic. Advancements in deep learning photography have made it easier for creators to use GANS in image processing, producing photos from text-based descriptions and upscaling real images. GANS can also be applied to simple retouching tasks like putting a smile on an unsmiling subject in a photo by “borrowing” pixels from an internal processing system.
Creating flawless images with machine learning is growing in popularity, making it increasingly difficult to tell original images apart from machine-generated ones. False images, sometimes known as deepfakes (in which a person in an existing image or video is replaced with someone else's likeness, for example) can be created with AI technology, and have gained notoriety for their use in spoofing news, creating satire and generally misleading people on whether an image is authentic or not. To help curb misinformation in this field, Adobe leads the Content Authenticity Initiative (CAI), which is a community of more than 350 media and tech companies, NGOs, academics, and others working to promote adoption of an open industry standard for content authenticity and provenance (record of ownership).
Storage and management
With the proliferation of smartphones, the number of photos we take increases daily, and it becomes increasingly difficult to organize them and find each one when we need them. It’s a challenge to sort through thousands of images no matter where you store them.
Adobe Lightroom users can catalogue all their photos and add keyword tags to find those images again. The machine learning model eliminates this tedious process by identifying saved images based on their distinct features. So you could type in a keyword like ‘wedding’ in the search bar and Lightroom will find all pictures of a wedding, or when making a composite with a blue background, type in ‘blue’ and it will find all the images that are predominantly blue in color.
AI, ML and creativity
AI has made it possible to enhance and manipulate images. It gives photographers and creators the ability to make more intelligent object selections and find images automatically based on what is in the photo itself rather than time-consuming searches. Machine learning can look at a photograph and decide how to best optimize it. The marriage of artificial intelligence and photography is constantly evolving and advancing and can help you hone your craft to create images, collages, media and other artwork you never imagined before.