Want Less Noise in Your Images? Add Semantic Information
Post · Apr 6, 2017 15:03 ·
When training an image classifier, one of the things we normally do is add noise to the dataset. This paper tackles the inverse problem (denoising) and adds semantic information to their images. Using ImageNet as their test set, the results are really compelling and show tons of promise / future research.
Highlights From the Paper
- Overcomes the regress-to-mean problem and enables the recovery of high frequency details.
- 3DR+AlexNet/VGG-16 do not overfit ImageNet images and generalizes well for color images outside this large dataset.
- In the case of σ = 25 , our proposed method 3DR+AlexNet gains 7.3% and 6.9% advantage over the case of classification without applying denoising for both AlexNet and VGG-16.
- Jiqing Wu
- Radu Timofte
- Zhiwu Huang
- Luc Van Gool
Large amount of image denoising literature focuses on single channel images and often experimentally validates the proposed methods on tens of images at most. In this paper, we investigate the interaction between denoising and classification on large scale dataset. Inspired by classification models, we propose a novel deep learning architecture for color (multichannel) image denoising and report on thousands of images from ImageNet dataset as well as commonly used imagery. We study the importance of (sufficient) training data, how semantic class information can be traded for improved denoising results. As a result, our method greatly improves PSNR performance by 0.34 - 0.51 dB on average over state-of-the art methods on large scale dataset. We conclude that it is beneficial to incorporate in classification models. On the other hand, we also study how noise affect classification performance. In the end, we come to a number of interesting conclusions, some being counter-intuitive.
Read the paper (pdf) »