Professional Documents
Culture Documents
Image Denoising Algorithm Based On PSO Optimizing Structuring Element
Image Denoising Algorithm Based On PSO Optimizing Structuring Element
Structuring Element
Zhu youlian, Huang cheng
College of Electronic Information Engineering, Jiangsu Teachers University of Technology, Changzhou, China, 213001
E-mail: czzhuyoulian@126.com
Abstract: A new image denoising algorithm is proposed to deal with information loss in the conventional morphological
image denoising process. The algorithm uses median operation to improve morphological operations’ performance,
which called median closing operation. It gives a mathematical model of the structuring element unit (SEU) composed of
a zero square matrix. The particle swarm optimization (PSO) algorithm is employed for choosing the size of structuring
element. The value of peak signal to noise ratio (PSNR) is taken as a fitness function, and the transformed value of the
particle's position is taken as the size of the structuring element. Experimental results show that the denoising
performance of the proposed algorithm has obvious superiority than conventional morphological algorithm. It can
overcome the inherent deficiency of conventional morphological operations, adaptively obtain the size of the structuring
element, and effectively remove impulse noise from images, especially for the image whose signal to noise ratio value is
relatively low. So it has a good prospect in image processing.
Key Words: Image Denoising, Morphological Filter, Particle Swarm Optimization (PSO), Structuring Element (SE);
Peak Signal-to-noise Ratio (PSNR)
978-1-4577-2074-1/12/$26.00 2012
c IEEE 2404
2 MEDIAN MORPHOLOGICAL 2.3 Denoising Properties of Different Median
OPERATIONS AND THEIR PROPERTIES Morphological Operations
Median morphological operations include median
2.1 Median Morphological Operations opening, median closing, median opening-closing, and
Suppose f is an image composed of a W × H set of median closing-opening. According to EQ. (5) and EQ. (6),
median opening f by SE is median erosion f by SE, followed
pixels, SE is referred to as structuring element (SE) in by median dilation of the result by SE; median closing f by
morphological operations, and it is composed of a m × n set SE is median dilation f by SE, followed by median erosion
of pixels. Then, two conventional morphological operations of the result by SE. Conventional erosion shrinks objects on
of f by SE can define as images, and conventional dilation expands them. So
{ }
Erosion: f ΘSE = min f − i , − j − SE (i, j ) , (1) median opening may expand objects on images, and
median closing shrink them. Impulse noise is suddenly
Dilation: f ⊕ SE = max{ f + SE (i, j )}.
i, j
(2) appeared discrete pulses with large amplitude and short
duration time, which should be better restrained by median
In erosion, the minimum value is selectedˈit can remove closing operation. Suppose we take Lena image and apply
image noise points that enlarged by noise, but can not different density noise to it. We respectively use four kinds
remove image noise points that reduced by noise. In of median morphological operations to remove noise from
dilation, the maximum value is selected and the visual Lena image. Some experimental results are shown in table
effect inverses to the eroded image. Details of the image are 1 and fig.1. According to fig.1, median closing operation
lost in erosion and dilation. Applications of conventional has obvious superiority than the others. Thus, the paper
morphology are restricted by extremum operations. Thus, adopts median closing operation to remove noise.
we define a new conception about median morphology in
which extremum operations of conventional morphology Table 1. Denoising properties of median morphological
are changed into median operation. Median erosion of f operations
by SE, denoted f ΘSE , median dilation of f by SE, Median Median
Median Median
SE’s
Noise Opening- Closing-
denoted f ⊕ SE , median operation denoted Median , they density
Closing Opening
Closing Opening
Size
PSNR(dB) PSNR(dB) n
are defined as PSNR(dB) PSNR(dB)
Median erosion: f ΘSE = Median{f −i ,− j − SE (i, j )}, (3) 30% 26.2661 25.7630 25.6455 25.6769 3
35% 25.1503 24.1089 23.6201 23.9110 4
Median dilation: f ⊕ SE = Median{f i, j + SE (i, j )} , (4)
50% 23.4201 22.8693 22.4391 22.3471 5
where i ∈ {1, 2 , " , m } , j ∈ {1, 2 , " , n } . Other median 60% 22.4661 21.3145 21.1131 20.9992 8
morphological operations are defined as
Median opening: f D SE = ( fΘSE ) ⊕ SE , (5)
Median closing: f • SE = ( f ⊕ SE )ΘSE , (6)
Median opening-closing: OC ( f , SE䯵 = f D SE • SE , (7)
Median closing-opening: CO( f , SE䯵 = f • SE D SE . (8)
5 SIMULATION EXPERIMENTS
The proposed algorithm is simulated in MATLAB7.0
and some experimental results are given in the below.
Experiment 1: comparison between conventional
morphological filter algorithm and the proposed algorithm. (n)
Fig. 2 Experimental results of the conventional morphological
Fig. 2 (a) is an original image of peppers. (b), (c) and (d)
filter and the proposed filter
are contaminated images with 15%, 30% and 55% density
impulse noise, respectively. (e), (f) and (g) are experimental Experiment 2: license plate number extraction.
results of the conventional morphological filtering
Fig. 3 (a) is a noisy image of 80% noise density. The
algorithm, respectively. (h), (i) and (j) are experimental
proposed algorithm is used to processing the noisy image
results of the proposed algorithm, respectively. (k), (l) and
and experimental result is shown in fig. 3 (b), its PSNR is
(m) are relations between the PSNR and iterations of the
equal to 19.6777.