An Improved HDRI Acquisition Algorithm
Based on Quality Measurement
Weihua Liu
1
, Shiqian Wu
2
1. School of Telecommunication and information
Engineering,
Xi'an University of Posts and Telecommunications,
Xi’an, 710121, P.R. China
hua.cat@163.com
Ying Liu
1
, Qian Wang
1
ˈNa Li
1
2. School of Machinery and Automation,
Wuhan University of Science and Technology,
Wuhan, 430081, P.R. China
shiqian.wu@foxmail.com
Abstract—In order to reduce the gap between dynamic range of
real scene and that of traditional imaging devices, a high dynamic
range (HDR) image acquisition algorithm based on quality
measurement is proposed. The algorithm fuses a stack of multi-
exposure image sequence in image domain. The weighting map of
an image is determined by a quality measurement, and then
further smoothed in order to reduce possible blocking artifact.
Experiments show that the proposed algorithm provides an HDR
image with rich details. Moreover, the HDR image acquired can
be clearly shown on low dynamic range displaying device without
tone-mapping work and save the complexity computation.
Keywords—high dynamic range image, image fusion,
exposedness, quality measurement , weighting map
I. INTRODUCTION
For a 8-bit standard imaging device, the pixel’s recorded
dynamic range is 0~255, which is much lower than that of real
scenes. As a result, a picture recorded is normally different
from what we perceive. High dynamic range (HDR) imaging
has been emerged to reduce the gap between the captured
images and the real scenes. The general method is to capture a
bracketed exposure sequence [1]. Image fusion is an efficient
solution to obtain an HDR image [2-4]. One is fusion in image
domain [5], i.e., the image sequence is fused into one by some
extra criteria achieved in image domain. The other is done in
radiance domain [6,7], i.e., the image sequence is first
transformed to radiance domain via camera response curve [8],
then an HDR image is generated in radiance domain followed
by tone mapping [9] to display in a standard device.
In this paper, we propose an improved fusion algorithm in
image domain based on detail character. The proposed method
has several advantages. First of all, the acquisition pipeline is
simplified, i.e., no in-between HDR image needs to be
computed. Second, the exposure information of each image is
not needed. The fine details are preserved better than that of the
algorithm of [14], and it can be implemented in real-time
application.
The paper is organized as follows. Section II reviews the
fusion algorithm in image domain and analyzes the effects of
the fused HDR image. Section III presents the improved-
weight-based fusion algorithm. The experimental results are
given in Section IV and conclusion remarks in Section V.
II.
REVIEWS OF THE FUSION ALGORITHM IN IMAGE DOMAIN
A. Existing Image Fusion Algorithms
Image fusion techniques have been used in many fields,
such as multi-focus image fusion [10], multi-spectrum image
fusion [11], image denoising based on fusion [12], video
enhancement [13], etc. In the early 90’s, Burt et al. [4]
proposed to implement fusion in a gradient pyramid. An
improved pyramid-based algorithm was presented in [14]. The
algorithm is more flexible by incorporating adjustable image
measures, such as contrast and saturation, and blends the
images in transform domain. The performance is good, but the
transform operator is complex. Goshtasby [15] also proposed
an algorithm to blend multiple exposures in time domain,
which partitions the image into uniform blocks and selects
each block that contains the most information. However, this
method cannot deal well with object boundaries and generates
blocking artifacts. Li et al. [18] proposed a detail-enhanced
fusion algorithm for multiple differently exposed images by
introducing a global weighted least square optimization
problem in gradient domain. However, ghost artifacts could be
an issue for the above image fusion algorithms. Li et al. [19]
introduced a ghost-free exposure fusion algorithm for multiple
differently exposed images with moving objects.
B. Features of Image Fusion Algorithm
Suppose the multi-exposure images are expressed as:
(, )= (, ) (, ), 1,2, ,
ll l
Yij X ij N ij l p+=" (1)
where, ( , )
l
Yijis an observed image, ( , )
l
Xijis the original
image, which is the ideally clear image, and ( , )
l
Nijis the
noise, which is independent and identically distributed (i.i.d.)
Gaussian white noise with mean 0 and standard deviation
0
σ
> , p is the number of captured LDR images. The fused
image is:
1
(, )* (, )
p
HDR l l
l
YwijYij
=
=
¦
(2)
Where the weighting maps satisfy:
1
(, ) 1, 0 (, ) 1
p
ll
l
wij and wij
=
=≤≤
¦
Feature 1: After the fusion, noise is reduced. This is shown as
below.
614
978-1-4799-8389-6/15/$31.00
c
2015 IEEE