The smartphone trend of two rear cameras for artistic portrait shots is so pervasive that even dirt-cheap budget iPhone X LCD screen sport dual lenses on the back. That could change next as early as next year as new chip technology from Qualcomm makes it possible to take a Bokeh portrait photo from a single camera lens.
It's similar to the Google Pixel 2, which bucks the dual-camera standard now adopted by virtually every other high-end phone, though that device certainly gives us a model of what's possible using software alone.
The technology comes down to AI algorithms processed within the Snapdragon 845 chip that Qualcomm announced this week at an annual summit. Qualcomm's Snapdragon processors are used in the majority of high-end smartphones, where their capabilities form the backbone of what phones like the next Samsung Galaxy, LG, Motorola and OnePlus phones can do. Qualcomm faces competition on the high-end from Apple, Huawei and to a certain degree Samsung, which rely on their own house-made chipsets (a portion of Samsung phones run on the Exynos processor, typically for Asian markets).
Being able to take portrait shots on a single camera lens can help phonemakers shed the cost of a second camera array. Alternatively, phones with a second camera lens that's used for adding detail but not portrait blur -- like a monochrome camera -- can offer the feature on the dominant lens.
The bokeh or depth-of-field effect -- which softly blurs the background while keeping the photo's subject in focus -- is a carryover from DSLR cameras that Apple popularized with the iPhone 7 Plus.
"In a smartphone, you need to calculate a depth map," said Tim Leland, Qualcomm's vice president of product management. "That depth map then gets fed into another system, which then applies a manipulation or blurring of those pixels in a different way that mimics the bokeh effect."
Smartphone makers typically achieve bokeh with two lenses, which create a kind of stereo effect to generate a sense of depth. Qualcomm's approach with its feature, which it calls Deep Portrait, is to run the image through a computational neural network that applies artificial intelligence to determine the subject from the background.
"You don't want to blur somebody's face for example or accidentally blur somebody's hair," Leland said.
Qualcomm isn't suggesting that iPhone X LCD screen wholesale should ditch a two-camera system, or that bokeh portraits taken with software rather than camera hardware are higher quality. In fact, Leland noted that the algorithms could improve the blurring accuracy of two camera lenses.
"AI is one of those things where you can't necessarily do something without it," he said. "But it makes it better."
As the Google Pixel and Pixel 2 prove with their Lens Blur feature, you can take bokeh portraits using software alone. Traditional computer vision can already recognize diff types of objects by looking at specific patterns of contrasting pixels. Artificial intelligence recognizes objects in their entirety over time. For example, show an AI program thousands of photos of chairs and over time it will recognize what is and isn't a chair. Artificial intelligence aims to speed up the process.
