Image denoising using deep convolutional autoencoder with feature pyramids

Ekrem Çeti̇nkaya,M Furkan Kiraç

doi:10.3906/elk-1911-138

Abstract

Image denoising is 1 of the fundamental problems in the image processing field since it is the preliminary step for many computer vision applications. Various approaches have been used for image denoising throughout the years from spatial filtering to model-based approaches. Having outperformed all traditional methods, neural-network-based discriminative methods have gained popularity in recent years. However, most of these methods still struggle to achieve flexibility against various noise levels and types. In this paper, a deep convolutional autoencoder combined with a variant of feature pyramid network is proposed for image denoising. Simulated data generated by Blender software along with corrupted natural images are used during training to improve robustness against various noise levels. Experimental results show that the proposed method can achieve competitive performance in blind Gaussian denoising with significantly less training time required compared to state of the art methods. Extensive experiments showed the proposed method gives promising performance in a wide range of noise levels with a single network.

Highlights

Image denoising is 1 of the fundamental problems in the image processing field due to being an essential step in many computer vision applications such as medical imaging
The results are compared with three state of the art methods, BM3D [3], DnCNN [4] and FFDNet [5]
Grayscale image denoising The proposed network can achieve competitive performance with state of the art methods and even surpasses BM3D [3] in BSDS68 dataset. It can achieve competitive denoising performance in Set12 dataset. It falls behind BM3D in this dataset and the possible main cause behind this is the repetitive structures of the images help BM3D to exploit nonlocal self-similarity

Summary

Introduction

Image denoising is 1 of the fundamental problems in the image processing field due to being an essential step in many computer vision applications such as medical imaging. Medical images tend to corrupt more if the radiation level is decreased [1, 2]. Denoising techniques are important to shift the balance towards less radiation exposure for patients in radiation level and image quality trade-off without sacrificing the image quality. The aim of image denoising is to obtain clean image x from corrupted version y that can be modeled as x = y + n where n is the noise of specific type. Most of the methods in the literature [3,4,5,6,7] focus on specific type for n , namely additive white Gaussian noise (AWGN) since natural images are assumed to have additive random noise which can be modeled with AWGN

Methods

Results

Conclusion