On the efficacy of deep image denoising for computer vision applications

Abstract

Image denoising is a process of inverse reconstruction where the original image is reconstructed from its noisy observations. Several deep learning models have been developed for image denoising. Usually the performance of image denoising is measured by metrics like peak signal to noise ratio (PSNR), structural similarity index (SSIM), however in this research we take a more pragmatic approach. We design and conduct experiments to evaluate the performance of deep image denoising methods in terms of improving the performance of some popular computer vision (CV) algorithms after image denoising. In this paper we have comparatively analysed: Fast and flexible denoising convolution neural network (CNN) (FFDNet), Feed forward denoising CNN (DnCNN) and Deep image prior (DIP) based image denoising. CV algorithms experimented with are face detection, face recognition and object detection. Standard and augmented datasets were used in our experiments. Raw images from standard datasets (BSDS500, LFW, FDDB and WGSID) were augmented with various kinds and levels of noise. From the results we obtained it can be concluded that image denoising is not effective in improving the performance of CV algorithms when denoising is applied to raw images of the datasets. But image denoising is very effective in improving the performance of the CV methods when denoising is applied to Gaussian noise corrupted images of the datasets. In our experiments we found results where the improvements were upto 11.70 percentage in terms of accuracy for face detection experiment.