Now you can snap DSLR quality photos using neural network

Nov. 1, 2017, 7:46 p.m. By: Pranjal Kumar

neural network image

Everyone love to click photos and there is no better than using DSLR. But we often find it very difficult to carry DSLR everywhere because of its size. This is the main reason why we invest heavily in our smartphone just to get a better camera. But, you would have realized that even the best smartphone camera cannot capture image of the same quality as of DSLR.

Yes, one of the way is that you can new Google Pixel phone and let the AI enhance your photos you press the shutter button. You can also use other Google’s services for minor fixes and clever effects. But, what if you want to have much better-quality image without using Google Pixel’s AI. Well, now you can do that using a neural network system that's focused solely on giving your photos a "DSLR-quality". The scientists have developed an approach doing that.

The researchers first trained a deep learning system using photos taken of the same scene using a phone and a DSLR. The network needs only two sets of images from different cameras to understand how to apply the image quality from one to the other. The basic principle is similar to taking the photo through one camera and then pass that image to the second camera. Yes, the result might not be great always but it will be certainly much better than what it was earlier. This is approach is best suited for the older or lower end devices.

But it has certain limitation also. It will not be able to add details that weren't already there. This means that if your device camera is not good at capturing low light images, then it will magically improve that.

The scientists did not stop there. They have also devised a way to change the shooting condition by using neural networks. It means that if you have clicked an image in a rainy day, you can make it appear like it was taken in the bright sunny day.

Scientist might have devised a new way to improve the quality of photos but yet they are little far from reaching the exact DSLR image.