Neural nets clean up digital snaps

Neural networks could boost digital image quality
Neural networks could boost digital image quality

Indian researchers have designed a neural network system that can remove noise and sharpen edges in digital images much more efficiently than current software.

Scientists at the Coimbatore Institute of Technology used a Modified Recurrent Hopfield Neural Network to reverse severe blurring and noise deliberately added to small (256-pixel) images.

The new inverse filtering system can quickly process an image, reducing distortion, noise and blurring, while using only limited computer resources.

Cleaning pixels one by one

Many image editors and photo cleanup software have built-in tools designed to remove noise and sharpen up edges. However, any clean-up process that works by changing individual pixels leads to degradation of the image and loss of information.

Earlier attempts at inverse filtering of an image relied on the image having a high signal-to-noise ratio. Other approaches require huge amounts of computing power.

The new neural net approach reduces information loss while reversing blurring caused by lens aberrations, and also reducing noise. Analysis shows that quality is improved by up to 67 per cent using the new approach, with results taking just half the time of less effective methods.

The scientists suggest that distortions in an image due to atmospheric disturbances could be unraveled and a photo taken on a hot, hazy day made acceptable. Because their neural network requires far fewer resources, it could also be built-in to cameraphones, boosting their notoriously poor image quality.