The Efficient-Parallel Stripe Noise Removal Algorithm with Low Resource
Utilization Based on FPGA
Jiao Chen, Hongxu Jiang
Beijing Key Laboratory of Digital Media
School of Computer Science and Engineering Beihang University
Beijing, China
e-mail: chenjiaobuaa@126.com, Jianghx@buaa.edu.cn
Abstract—Stripe noise is usually introduced into infrared
images, which seriously deteriorates the image quality and
affects the subsequent analysis of the image. The main
formation causes of stripe noise are analyzed and some
typical stripe noise removal methods and their limitations
are compared. On this basis, this paper presents a suitable
method for FPGA implementation stripe noise removal in
infrared image, i.e. based on the most stable window strip
noise removal method. The proposed method mainly uses
the average compensation method principle and the relevant
features between adjacent rows of pixels to implement. And
it applies the average difference between line represent the
noise difference and takes advantage of FPGA parallel
to optimize. Experimental results indicate that the proposed
method not only can remove stripe noise completely and
assure won’t produce artifacts, but also can realize efficient
parallel operation in the low resource utilization way.
Compared with the traditional hardware method, the
de-noising effect is better, and the resource utilization is the
lowest. Compared with the typical software method, it
realizes the algorithm of parallel computing and the speed of
the proposed algorithm has a big improvement.
Keywords—strip noise; FPGA; average compensation;
parallelism; resource utilization; infrared image
I. INTRODUCTION
Because of the limitation of detector material and
manufacturing process, response degree of each detection
unit for the same irradiation is not completely same on the
focal plane. Therefore, using linear array of scanning
detector always leads to depth and white horizontal stripes
ribbon noise
[1]
in infrared image. Among the line integral
readout circuit, this leads to the special noise owing to the
non-uniformity of readout circuit and bias voltage noise,
which is also called stripe noise
[2]
. It seriously affects the
image quality of the infrared system, and the noise
parameters will slowly drift with the changes in the
external environment. When the image is enhanced, the
stripes become more prominent, and even show the shape
of the shutter. Therefore, how to effectively remove stripe
noise to realize the non-uniform correction is the key to
improve the quality of infrared images.
Noise pollution has a direct impact on image edge
detection, feature extraction, image segmentation, and
target detection et al. Therefore, the use of appropriate
methods to eliminate noise is a very important
pre-processing step in image processing. Now image
processing technology has been deep into the scientific
research, military technology, industrial and agricultural
production, medicine, meteorology, astronomy and other fields.
The scientists can use satellite photos to obtain the earth's
resources and weather situation; Doctors can analyze the
tomography images of various parts of human body by X-ray
or CT. But because of the existence of stripe noise, it not only
affects the image visual effect, also reduces the ratio of
signal-noise and the precision of subsequent data analysis.
Therefore, it is necessary to study the effective algorithm of
removing stripe noise.
This article proposes the new method only by using a pair
of image output,
i.e. based on the most stable window stripe
noise removal method, which can remove stripe noise
completely and is suitable for real-time processing of the
image.
II. R
ELATED WORK
On the software, the causes of formation stripe noise and
the denoising method have been studied by many scholars. To
date, the methods used to remove stripe noise are divided
mainly into two categories. The One is using the filter operator
to remove frequency components of periodic noise through
some transform in frequency domain. The defect is not easy to
choose the right frequency components. Another is according
to the image grey value characteristics normalized (standard)
and matched method. The typical methods have histogram
matching, moment matching method
[3]
, et al. Its drawbacks,
however, is to change the original reflectance distribution
characteristics and often requires a sufficiently large image and
has a uniform distribution.
The method of space-frequency filtering
[4]
is suitable for
geometric correction of remote sensing image, and has a good
effect in some special cases. But this method has some
shortcomings. On the one hand, it needs three steps: transform,
filter processing and inverse transform, and needs complex
computing process and high computing requirements; On the
other hand, it is difficult to set a suitable filter processing, and
the process needs to trial and error, resulting in difficult to
achieve batch; In addition, under the distribution features of
complex surface, stripe noise and image texture themselves are
mixed. Removal of stripes will also destroy the image details,
thereby reducing image quality.
In the filter method of a local adaptive threshold noise
detection linear interpolation filter (LALIF) for stripe noise
removal
[5]
has a good performance in adaptability and stability
for complex stripe noise removal in infrared images. However,
2017 IEEE 2nd International Conference on Signal and Image Processing
978-1-5386-0969-9/17/$31.00 ©2017 IEEE