Multidimensional Systems and Signal Processing
https://doi.org/10.1007/s11045-019-00655-6
Fusion of multi-exposure images using recursive
and Gaussian filter
Vishal Chaudhary
1
· Vinay Kumar
1
Received: 5 March 2018 / Revised: 7 May 2019 / Accepted: 10 May 2019
© Springer Science+Business Media, LLC, part of Springer Nature 2019
Abstract
This paper proposes a novel technique to create a high-resolution image by combining
the bracketed exposure sequence without a priori knowledge of source image. The source
image is split into three categories: constant, high varying and low varying feature images.
For high and low varying features, pixels with highest information is selected and combined
to construct collective high and low varying feature image. Collective constant feature image
is constructed from weighted average of constant feature images, where weight is calculated
based on information present in original source images. These pre-processed high, low and
constant feature images are further combined to produce a final fused image. Objective anal-
ysis based quality evaluation parameters show a significant improvement in result produced
by proposed method against the state-of-the-art.
Keywords Dynamic range · Exposure time · Recursive filter · Image fusion
1 Introduction
A scene varies largely by exposure level, therefore a single LDR (low dynamic range) camera
can never collect all underlying informations of the scene. The captured image may experience
under or over-exposed areas due to high or low shutter speed of the camera, respectively. High
shutter speed or low exposure time results in information loss in darker region and vice versa
(refer Fig. 1). An image constructed by collecting complementary information from varying
exposure time LDR images (source images) can preserve all the details of the scene. Two
major approaches to perform the operation are: HDR (high dynamic range) tone mapping
and multi-exposure image fusion. LDR source images are combined together to construct
HDR tone mapped image by extracting Camera Response Function (CRF). Extraction of
CRF requires additional information of source images; like, exposure time and exposure
value. Next, dynamic range of resulting HDR image is compressed to LDR while preserving
B Vishal Chaudhary
vishalch.nsit@gmail.com
1
Department of Electronics and Communication Engineering, Thapar Institute of Engineering and
Technology, Patiala, India
123