Citation: Guo, Y.; Lu, W.; Li, X.;
Huang, Q. Single Image Reflection
Removal Based on Residual
Attention Mechanism. Appl. Sci. 2023,
13, 1618. https://doi.org/10.3390/
app13031618
Academic Editors: Phivos Mylonas,
Katia Lida Kermanidis and
Manolis Maragoudakis
Received: 29 November 2022
Revised: 24 January 2023
Accepted: 24 January 2023
Published: 27 January 2023
Copyright: © 2023 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
Single Image Reflection Removal Based on Residual
Attention Mechanism
Yubin Guo
1,2
, Wanzhou Lu
1,2
, Ximing Li
1
and Qiong Huang
1,2,
*
1
College of Mathematics and Informatics, South China Agricultural University, Guangzhou 510642, China
2
Guangzhou Key Laboratory of Intelligent Agriculture, Guangzhou 510642, China
* Correspondence: qhuang@scau.edu.cn
Abstract:
Affected by shooting angle and light intensity, shooting through transparent media may
cause light reflections in an image and influence picture quality, which has a negative effect on the
research of computer vision tasks. In this paper, we propose a Residual Attention Based Reflection
Removal Network (RABRRN) to tackle the issue of single image reflection removal. We hold that
reflection removal is essentially an image separation problem sensitive to both spatial and channel
features. Therefore, we integrate spatial attention and channel attention into the model to enhance
spatial and channel feature representation. For a more feasible solution to solve the problem of
gradient disappearance in the iterative training of deep neural networks, the attention module is
combined with a residual network to design a residual attention module so that the performance of
reflection removal can be ameliorated. In addition, we establish a reflection image dataset named the
SCAU Reflection Image Dataset (SCAU-RID), providing sufficient real training data. The experimental
results show that the proposed method achieves a PSNR of 23.787 dB and an SSIM value of 0.885
from four benchmark datasets. Compared with the other most advanced methods, our method has
only 18.524M parameters, but it obtains the best results from test datasets.
Keywords:
artificial neural network; image processing; image restoration; computer vision; artificial
intelligence; supervised learning; multi-layer neural network
1. Introduction
Images captured through glass have a noticeable layer of reflection due to the shooting
angle and light intensity. Image reflection can reduce the image quality and adversely
affect the results of computer vision tasks, such as image classification and object detection.
Accordingly, reflection layers are expected to be removed to obtain clear images.
In this study, the research objective of image reflection removal is to predict the
transmission layer T from a given reflection image I. According to [
1
], let I be the reflection
image, T the transmission layer and R the reflection layer, then the reflection image can be
approximately modeled as a combination of T and R. It can be seen that to any I, both the
transmission layer T and the reflection layer R are unknown. As there are no additional
constraints or priors, the equation has infinite feasible solutions.
To solve this problem, it is imperative for researchers to impose constraints and
artificial priors on the solution space, thus a separation of the reflection image closer to
an ideal target solution can be obtained. As for the ill-posed problem, [
2
] proposed the
concept of relative smoothness for reflection image separation. That is, the reflection layer
is considered to be smoother relative to the transmission layer, so a smooth gradient is
applied to the objective function of the reflection layer, and a sparse gradient is applied
to the objective function of the transmission layer. There are other solutions proposed.
For example, [
3
] introduced the use of ghosting cues that exploit the asymmetry between
layers, thus helping to reduce the discomfort of eliminating reflections in images taken
through thick glass. The authors of [
4
] proposed a simple yet effective reflection-free cue
Appl. Sci. 2023, 13, 1618. https://doi.org/10.3390/app13031618 https://www.mdpi.com/journal/applsci
