Proceedings of the 2nd International Conference on Computer Science and Electronics Engineering (ICCSEE 2013)

Comparison of Commonly Used Image Interpolation Methods

Dianyuan Han
Dept. of Computer Engineering
Wei Fang University
Shandong 261061, China
wfhdy@163.com

Abstract—Image magnification algorithms directly affect the such as P. so the values of these pixels should be calculated
quality of image magnification. In this paper, based on the throw interpolating estimation.
image interpolation algorithm principle, features of the nearest
neighbor interpolation, bilinear interpolation, A B A B
bicubic interpolation and cubic B spline interpolation were
analyzed. At the same time, their advantages and C D
disadvantages were compared. In the experiment, image
magnification performance of different interpolation
algorithms was compared from subjective and objective P
aspects. The experimental results give the guidance for the user
to choose a suitable algorithm to achieve optimum results C D
according to different application.

Keywords- Image magnification; Interpolation algorithm; (a) (b)

Performance comparison
Figure 1. Schematic diagram of image magnification

I. INTRODUCTION
Image magnification is one of the basic image operations, III. COMMON INTERPOLATION ALGORITHM
and is widely used in many applications[1]. The image
magnification is a conversion process from a low resolution A. Nearest neighbor interpolation
image to a high resolution image. The image magnification is In nearest neighbor interpolation algorithm, the position
essentially image interpolation process. There have been a of pixel P in the magnified image is converted into the
lot of practical image magnification methods which have original image, and the distance between P and its neighbor
their own characteristics, advantages and disadvantages. The points A, B, C and D were calculated. Then the color values
choice of the different image magnification algorithm of pixel P was set as the values of the pixel which was
directly affects enlarged image quality, so it is important to nearest to P[2].
find a suitable algorithm to improve the image magnification In Fig. 2, suppose ( i, j ) , ( i, j + 1) , ( i+ 1, j ) and ( i+ 1,j
quality. + 1) are the 4-neighbor points, and there values are f( i, j ) ,
In current, the commonly used image magnification f( i, j + 1) , f( i+ 1, j ) and f( i+ 1,j + 1). The distance between
method is interpolation method. In the experiment, image (u,v) and ( i, j ) , ( i, j + 1) , ( i+ 1, j ) and ( i+ 1,j + 1) were
magnification performance of different interpolation calculated, then the values of (u,v) was set as the value of
algorithms was compared from subjective and objective the point which is nearest to (u,v). [3]
aspects to give the guidance for the user to choose a suitable
algorithm to achieve optimum results according to different
(i,j) (i,j+1)
application.
II. IMAGE AMPLIFICATION PRINCIPLE
(u,v)
The basic principle of image magnification is to increase
the image pixel number, so a low resolution image is
converted to a high resolution image. (i+1,j) (i+1,j+1)
When a small image is enlarged, for example if an image
is zoomed to 400% shown in Fig. 1., the color values of Figure 2. Diagram of nearest neighbor interpolation algorithm
original 4 adjacent pixels marked A, B, C, and D in (a) were
filled in the new A, B, C, and D locations in (b) accordance
with the magnification factor. But there are a large number B. Bilinear interpolation
of pixels which values are unknown between A, B, C, and D, In bilinear interpolation algorithm[4], the position of
pixel P in the magnified image was converted into the

Published by Atlantis Press, Paris, France.

© the authors
1556
 Proceedings of the 2nd International Conference on Computer Science and Electronics Engineering (ICCSEE 2013)

original image, then the Influence of the four pixel points A, nonlinear calculation. The diagram of nonlinear interpolation
B, C and D was calculate. The nearer distance to the point P, algorithm is shown in Fig.5.
the value is greater, which indicates the greater effects. The
diagram of bilinear interpolation is shown in Fig. 3.

A E B

Figure 5. Diagram of nonlinear interpolation algorithm

P
On the basis of calculating the lateral interpolation effects,
C F D the operation as mentioned above is extended into two
dimensional space to calculate the longitudinal interpolation
Figure 3. Diagram of bilinear interpolation algorithm effects.

Suppose the coordinate of A, B, C and D are ( i, j ) , ( i, j D. Cubic B-SPline

+ 1) , ( i+ 1, j ) and ( i+ 1,j + 1). The coordinate of P is (u,v). In order to get better image quality, many new algorithms
The bilinear interpolation algorithm consists of three steps: have been proposed, they use more complex improved
Step 1: to calculate the influence of A and B and denote it interpolation method such as B spline [5] and Mitchell
as E. interpolation algorithm. Their objective is to make the
f (i, j + v) = [ f (i, j + 1) − f (i, j )]v + f (i, j ) (1) interpolation curve more smoothly and the image edge more
Step 2: to calculate the influence of C and D and denote it perfect.
as F. The function of cubic B-spline is as follow[2].
f (i + 1, j + v) = [ f (i + 1, j + 1) − f (i + 1, j )]v + f (i + 1, j ) +∞

(2) f ( x) = B
k = −∞
k , n +1 ( x) • f ( x k ) (4)
Step 3: to calculate the influence of E and F and denote it
as P. Three order B-spline function is as follows:
f (i + u, j + v) = (1 − u )(1 − v) f (i, j ) − (1 − u )vf (i, j + 1)  ( x − xi ) 2
 ，xi ≤ x ≤ xi+1
( xi +1 − xi )( xi +2 − xi )
+ u (1 − v) f (i + 1, j ) + uvf (i + 1, j + 1) 
 ( x − xi ) 2 ( xi +2 − x)( x − xi +1 )
(3)  − ，xi＋1 ≤ x ≤ xi+ 2
Bi ,3 =  ( xi +1 − xi )( xi + 2 − xi ) ( xi + 2 − xi +1 )( xi +3 − xi +1 )
C. Bicubic Interpolation  ( x − xi + 3 ) 2
 ，xi＋2 ≤ x ≤ xi +3
Bicubic interpolation is similar to bilinear interpolation  ( x i +3 − xi +1 )( xi +3 − xi + 2 )
algorithm. For the unknown pixel P in amplified image, its  0 everywhere
influence sphere is expanded to its 16 adjacent pixels, then (5)
the color value of P is calculated by these 16 pixels
according to their distance to P. The diagram of bicubic
interpolation algorithm is shown in Fig.4. IV. EXPERIMENT RESULT
In order to test the performance of the algorithm, We
(i-1, j-1) (i-1, j+2) have developed a MATLAB 2009[6] implementation on
Intel(R) Core(TM) 2 T7200 2.00GHz CPU, 2.0GB RAM
(i, j)
computer.
(u, v)
To compare the performance of the different algorithm, a
test image was zoomed out to half of the original size. And
then the shrunken image was expanded to its original size by
(i+2, j-1) (i-2, j+2)
using different algorithm.
At last the enlarged image quality was compared From
two aspects: subjective and objective.
Figure 4. Diagram of bicubic interpolation algorithm A. Objective Evaluation
In this paper, the mean square signal-to-noise ratio (SNR)
Compared with bilinear interpolation algorithm, bicubic was used to evaluate image expansion performance of
interpolation algorithm extends the influence with more different algorithms.
points, and uses advanced interpolation algorithm. Let f (x, y) represents the original image with the size of
In the horizontal direction, to calculate the value of e 
M × N, f ( x, y ) represents the enlarged image which has
between two points A and B, it is need to use four pixels
been zoomed out to half size of its original image, and
value of A, B , A-1, B+1 and get a smooth curve through a

Published by Atlantis Press, Paris, France.

© the authors
1557
 Proceedings of the 2nd International Conference on Computer Science and Electronics Engineering (ICCSEE 2013)

f represents the mean gray value of the original image. f Average Running Time
is defined as fellow.
M −1 N −1
1
 f ( x, y) 18 17
f = (6)
16
MN x =0 y =0
14
SNR is defined as (7). 12
 M −1 N −1  10
  [ f ( x, y) − f ]
2 8
 8 6
 x =0 y =0  6 5
SNR = 10 lg M −1 N −1  (7)
 4
  [ f ( x, y ) − f ( x, y)]2  2
 x =0 y =0  0 Neighbour B-SPline
In the experiment, we selected 30 images representative Nearest Bilinear Bicubic Cubic
image to test, the results is shown in Tab.1.

TABLE I. Contrast table of SNR with different interpolation mothods Figure 6. Average running time of different interpolation algorithm

Tested Nearest Bilinear Bicubic Cubic

Image Neighbour B-SPline B. Subjective Evaluation
image 1 19.1112 23.3023 23.4204 22.5028 Image is used to be watch by people, so it is more
image 2 16.0059 18.4308 18.5899 17.9319 suitable evaluate the magnified image quality with subjective
image 3 20.0614 25.3736 25.6634 24.3505
methods.
In this paper, 20 observers were invited to observe
image 4 15.9497 18.7056 18.9392 18.1074
magnified images and give evaluation. At last the image
image 5 17.1252 20.2387 20.9872 18.6582 scoring average and used to evaluate the subjective image
image 6 18.15564 22.137185 22.24938 21.37766 quality.
image 7 14.40531 16.58772 16.73091 16.13871 The subjective evaluations with different interpolation
image 8 17.05219 21.56756 21.81389 20.697925 methods are shown in Tab. 2.
image 9 12.75976 14.96448 15.15136 14.48592
TABLE II. Contrast table of subjective evaluation with different
image 10 12.8439 15.179025 15.7404 13.99365 interpolation mothods

Interpolation Subjective Image Overall Processing

SNR can reflect the magnified image quality, the larger Type Feelings Contour Evaluation Time
SNR means the higher image quality. (seconds)
It can be seen from the above table that the SNR between obvious
Nearrst_
the original image and the magnified image get from the mosaic Not clear Worst 5
Neighbour
phenomenon
nearest neighbour interpolation is the minimum, and bilinear not clear,
interpolation algorithm is much larger, bicubic interpolation blur, not
Bilinear serrate Poor 6
is the largest, and the B spline larger than the nearest sharp
phenomenon
neighbor method but less than the bilinear interpolation serrate
algorithm and bicubic interpolation method. fuzzy, phenomenon
Bicubic Better 8
sharper has
The complexity of Different interpolation algorithm is improved
different, and their average running time also has bigger Edge
difference. The average run time is shown in Fig.6. becomes
relatively
B-Spline clear, serrate Good 17
clear, sharp
phenomenon
disappeared

An example of image magnification is shown in Fig.7.

Published by Atlantis Press, Paris, France.

© the authors
1558
 Proceedings of the 2nd International Conference on Computer Science and Electronics Engineering (ICCSEE 2013)

interpolation algorithm is continuous, so the visual effect is

better than the nearest point interpolation algorithm, but the
operation speed is slightly slower. Because of this, if
someone pays more attention to speed and good visual effect,
this is a good compromise.
Bicubic interpolation algorithm can get relatively clear
picture quality, but it needs larger amount of calculation.
This algorithm now is most commonly used in many image
processing software such as Photoshop, After Effects, Avid
(a)Original image(256×256) (b) Shrunken image(64×64)
and Final Cut Pro etc.
The magnified image by using of cubic B-spline
interpolation algorithm is smooth and has no obvious saw
tooth phenomenon. But in the interpolation process it takes
the form of a low pass filter and suppress high frequency
component in varying degrees. When the amplification
factor is higher, it will cause the edge fuzzy and false
artificial traces.
The experimental results give the guidance for the user to
choose a suitable algorithm to achieve optimum results
according to different application.
(c) Nearest neighbor Interpolation (d) Bilinear interpolation
ACKNOWLEDGMENT
We acknowledge the financial support of Weifang
science and technology development program
(NO.20111127) and Shandong province science and
technology development program (NO.2011YD01047).
REFERENCES
[1] van Ouwerkerk J D. “Image super resolution survey”. Image and
Vision Computing, 2006, 24( 10) : 1039-1052.
(e) Bicubic interpolation (f) Cubic B-SPline interpolation [2] Jia Yonghong. “Digital image processing(The Second Edition)”.Wu
Han China: Wu Han university press, 2010: 114-116.
Figure 7. Diagram of nonlinear interpolation algorithm
[3] Zhao haifeng, Zhou Yongfei, Huang Ziqiang. “Comparison of Image
Amplifying Method”. Modern Electronics Technique, 2010,(24):33-
36.
V. CONCLUSIONS
[4] Wang Sen, Yang Kejian. “An image scaling algorithm based on
Nearest neighbour interpolation algorithm is the most bilinear interpolation with VC+ +” . Jou rnal of Techniques of
simple and fast algorithm. It has the advantages of fast speed, Automation & Applications, 2008, 27( 7) : 44-45.
but it can bring significant distortion and it will appear [5] Feng Jiefei1, Han Huijian.”Image enlargement based on non-uniform
mosaic and saw tooth phenomenon. B-spline interpolation algorithm”. Journal of Computer Applications,
Bilinear interpolation method is more complex than the 2010,30(1):82-84.
nearest neighbour method, and so it has larger calculation. It [6] Rafael c. Gonzalez, Richard E. Woods. “Digital image processing
second Edition”, Beijing: Publishing house of electronics
has no gray discontinuity defects and has satisfactory results. industry,2008,.
This method has a low pass filtering properties, so that the
high frequency component is faded and the image contour
has some degree of fuzzy. As the results of bilinear

Published by Atlantis Press, Paris, France.

© the authors
1559