Researchers at NVIDIA, Aalto University in Finland, and MIT have together discovered a new technique to fix noisy or grainy photos using a deep-learning algorithm. The idea is being presented at the International Conference on Machine Learning (ICML) 2018 in Stockholm, Sweden this week. Photos taken in low light are often filled with noise. Grainy or pixelated images are not exactly what people desire from their photography work. In such a case, deep-learning algorithms that fall under the broader concept of machine and artificial intelligence can come handy as this particular paper revealed.
In the research paper dated March 18, 2018, researchers have explained the math and science behind this deep-learning based approach, which aims to train a neural network to fix images by feeding example pairs of noisy and clean images. Meaning, initially, it requires as many examples and training sets as possible. Later on, it recognises and restores images on its own.
Thanks to the artificial intelligence (AI), it learns to differentiate between noisy and noise-free images. According to researchers, this technique is different from the traditional ones given the fact that it only requires two input images with the noise or grain. Once trained properly, the AI can fix noise or grain and enhance photos without being shown what noise-free images look like.
The team of researchers trained their system, courtesy of 50,000 images in the ImageNet validation set. It could be achieved with the help of NVIDIA Tesla P100 GPUs with the cuDNN-accelerated TensorFlow deep learning framework.
“We show applications in photographic noise removal, denoising of synthetic Monte Carlo images, and reconstruction of MRI scans from undersampled inputs, all based on only observing corrupted data,” the researchers stated in their paper. “Our proof-of-concept demonstrations point the way to significant potential benefits in these applications by removing the need for potentially strenuous collection of clean data,” they further added.