Image enhancement for smartphones

Technik -

Image enhancement for smartphones

New image enhancement model for smartphone real-time applications

The latest camera designs have seen major advances in image processing pipelines. These enable the capture of increasingly high-quality images on smartphones. However, one thing remains: The sensor is simply too small to keep up with the large sensors of compact or DSLR cameras in terms of quality due to the form factor. The lenses are also physically too small. Therefore, the recording quality cannot match the quality of larger cameras for physical reasons alone.

Nevertheless, it's amazing what image quality the latest smartphone camera models achieve. However, artifacts and other impairments in the processed images are very common with smartphone cameras. Noise artifacts are the most common and diffraction artifacts, HDR overexposure and blur are also typical problems that smartphone camera models have to deal with.

Now, for the first time, a team from Computer Vision Lab (Marcos V. Conde, Florin Vasluianu, Javier Vazquez-Corral, Radu Timofte) has successfully removed these artifacts in image restoration using deep learning methods. They used LPIENet, a lightweight network for perceptual image enhancement, which is mainly intended to be used on smartphones.

"Our experiments show that, with much fewer parameters and operations, our model can deal with the mentioned artifacts and achieve competitive performance compared with state-of-the-art methods on standard benchmarks."

The efficiency and reliability of this approach was directly proven on normal commercial smartphones and the performance was evaluated. The developed model was able to process images with 2K resolution in less than 1 second on standard mid-range smartphones. This performance thus allows the images to be processed with only a slight delay on the smartphones.

It remains to be seen when we will see these technologies offered on smartphones. However, one thing will not change: the form factor. And that brings us back to the main factors that limit the image quality. Such small sensors still can't keep up with the "big ones", but maybe AI will soon allow us to break these limits.

What cannot be "blown up" so easily, however, is the difference in the imaging quality of the lenses used, although these already deliver amazing results on smartphones. Nevertheless, an elephant will probably not fit through the famous keyhole for physical reasons.

 


Leave a comment