See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/274702043

Zooming Digital Images using Modal Interpolation

Article · May 2013

CITATIONS READS

2 222

1 author:

Robin Walia
Maharishi Markandeshwar University, Mullana
9 PUBLICATIONS 20 CITATIONS

SEE PROFILE

All content following this page was uploaded by Robin Walia on 09 April 2015.

The user has requested enhancement of the downloaded file.

International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: [email protected], [email protected]
Volume 2, Issue 5, May 2013 ISSN 2319 - 4847

Zooming Digital Images using Modal

Interpolation
SUDHIR SHARMA1, Asst. Prof. ROBIN WALIA2
1&2
Department of Electronics & Communication Engineering,
Maharishi Markandeshwar University,
Mullana (Ambala), INDIA

ABSTRACT
Image zooming is the process of enlarging an image by a factor (greater than one) called magnification. Image zooming
involves re-sampling and interpolation which cannot be ideally performed infinitely. Therefore, when applying zooming
function indiscriminately to an image, it will generally result in aliasing; edge blurring and other artifacts. This paper
considers two traditional interpolation schemes (nearest neighbor & bilinear) and makes an adaptive method using these two.
This is an adaptive algorithm and is based on assuming that one of the four pixels has a value that is appropriate enough to be
assigned directly at an empty location. In this paper, the statistical mode parameter is used to select this value from the four
pixels. The results have been evaluated in terms of visual appearance and PSNR. From the results, it is observed that
depending on the interpolation ratio selected or set (i.e. depending on the final size desired/targeted), the interpolation
algorithms gave different PSNR as well as visual quality. It is observed that the proposed algorithm gives comparatively higher
values of PSNR and better visual appearance.

Keywords: Zooming, Resizing, Interpolation, magnification.

I. INTRODUCTION
A digital image is build up of elements called pixels. Mathematically, it can be defined as a two dimensional function
f(x, y), where x and y are spatial (plane) co-ordinates. The value of ‘f’ at any pair of co-ordinate (x, y) is called the
intensity ‘or’ gray level of the image at that point. The processing of digital images is called digital image processing.
The most common digital image processing tasks include: resizing, zooming and color enhancement. This paper
focused on zooming of digital images. Image zooming means changing the number of display pixels per image pixel
only in appearance. For zoom level= 1, there is one display pixel per image pixel. However for zoom level= 2, there are
2 display pixels per image pixel in both x and y axis. This enlargement is usually quantified in terms of a number
(greater than one) called magnification. In terms of applications, the process of image zooming is usually applied in
diverse areas ranging from computer graphics, rendering, editing, medical image reconstruction, to online image
viewing.
The zooming operation is performed using resampling and interpolation operations. Traditional image zooming
techniques use up-sampling by zero-insertion followed by one-dimensional filtering to interpolate the high resolution
(HR) samples [1-3]. The main drawback of this approach is that the frequency content of the high-resolution image is
same as the low-resolution (LR) images. This is due to the fact that linear techniques are incapable of introducing new
information into the image. The lack of new high frequency content results in a variety of undesirable image artifacts
such as blocking, staircase edges and blurring. This paper, therefore, presents an image zooming algorithm based on
the combination of the traditional non-adaptive methods: nearest neighbor and bilinear interpolation methods. The
nearest neighbor algorithm does not permit to create a new value, instead set the value at the empty location by
replicating the pixel value located at the shortest distance [4-5]. The effect of this is to make image pixel bigger which
results in heavy jagged edges. A solution to such jaggedness was achieved through the bilinear interpolation [6-8]. A
bilinear based algorithm generates softer images but blurred thus making the algorithm inappropriate also for HR
applications. The blurredness problem was reduced by introducing the convolution based techniques [9-11]. Such
algorithms performed better than the two in terms of the visual quality but are inappropriate to use where the speed is
of the prime importance. Now, as the source image resolution is often reduced after undergoing the interpolation
process, the easy way to generate a HR image using linear interpolation means is to reduce, at any cost, any operation
that would underestimate or overestimate some parts of the image. This can be done by assuming that one of the four
pixels has a value that is appropriate enough to be assigned directly at an empty location. In this paper, the statistical
mode parameter is used to select this value from the four pixels. The results have been evaluated in terms of visual
appearance and PSNR.

Volume 2, Issue 5, May 2013 Page 305

International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: [email protected], [email protected]
Volume 2, Issue 5, May 2013 ISSN 2319 - 4847

II. SYSTEM MODEL

In zooming operation, the re-sampling and interpolation operations are used. The principle of image zooming is shown
below in Figure 1. In Figure 1(a), the image A is magnified to 2 times to image B. The black circles are the unknown
pixel values. The image A (2x2) is magnified to B (4x4); however in reality, the original image (nxn) is magnified to
(2n-1)x(2n-1). This process of zooming includes three stages as reported in Figure 1.

a) Magnifying an image b) Stage I

c)

c) Stage II d) Stage III

Figure 1: Principle of zooming an image.

A. Stage I:
The first step is to expand the original (nxn) image to (2n-1)x(2n-1). This can be done by extending the initial image
borders. For example in Figure 1(b), the X(5x5) is original image and Y(9x9) is the magnified image. X(i,j) is the pixel
in original image; where i is the ith row and j is the jth column. In the same way in Y(m,n), m is the mth row and n is
the nth column. The black dots represent the pixels in the original image (X) and these pixels can be mapped in image
(Y) as X(i,j)=Y(2i-1,2j-1). The white dots represent the unknown pixels. When an image is magnified 2 times, total
number of pixels will be four times; thus, we have now 1 known and 3 unknown pixels.
B. Stage II:
In this stage, the centre pixel X (Figure 1(c)) has to be finding out. The centre pixel X is deduced with the help of
proposed algorithm in the next section. This procedure is repeated for the complete image until all the unknown pixels
of the boundaries and the centre pixels are found.
C. Stage III:
In this stage, we have to start from the beginning of the image and find the left over pixels as shown in Figure 1(d). In
Figure 1(d), A & B are the pixels from original image, and X1 and X2 are the pixels derived from stage I. All the pixels
(A, B, X1, X2) are considered as in Stage I and computed in the same way. Finally, the value of M is computed and put
in the place of centre pixel.
From the above discussion, when performing a digital image interpolation, we are actually creating empty spaces in the
source image and filling in them with the appropriate pixel values. This makes the interpolation algorithms yielding
different results depending on the concept used to find those values. For example, in the nearest neighbor technique,
the empty spaces will be filled in with the nearest neighboring pixel value. This (nearest neighbor algorithm) concept is
very useful when speed is the main concern. Unlike simple nearest neighbor, other techniques use interpolation of
neighboring pixels, resulting in smoother image. A good example of a computationally efficient basic re-sampling
concept or technique is the bilinear interpolation. Here, the key idea is to perform linear interpolation first in one
direction, and then again in the other direction. Although each step is linear in the sampled values and in the position,
Volume 2, Issue 5, May 2013 Page 306
International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: [email protected], [email protected]
Volume 2, Issue 5, May 2013 ISSN 2319 - 4847

the interpolation as a whole is not linear but rather quadratic in the sample location. In other words, the bilinear
interpolation algorithm creates a weighted average value that uses to fill in the empty spaces. This provides better
trade-off between image quality and computational cost but blurs the interpolated image thus reducing its resolution.
This paper uses a combination of these two algorithms to find where to apply nearest neighbour and bilinear.

Proposed Method
The proposed method is shown in Figure 2, which comprises of four known pixels (A, B, C, D) and the fifth unknown
pixel (E). To determine this unknown pixel (E) the algorithm is shown in Figure 3 and detailed below.

Figure 2: Four neighbor locations around an empty location E.

The first step is to compute the mode of the four nearest neighbor pixels. Consider Figures 2 & 3, in which the four
pixels A, B, C, D are known and E is the unknown pixel. These four known pixels are stored in an array DS=[A, B, C,
D] and the mode of this set is calculated. The term ‘mode’ in statistics means the value that occurs most frequently in a
data set. If a mode exits in the data set (DS), then the algorithm applies the nearest neighbor, i.e., assigns the mode
value (nearest neighbor) to that empty location. If the mode does not exist in DS (i.e., when two data in DS appear the
same number of times or when none of the DS data repeat) then, the bilinear interpolation in performed on the DS data
to determine the value of the empty new pixel (E). This is done as follows.
The bilinear value (BV) is subtracted from each pixel value in date set (DS), as:
NDS=|DS-BV| (1)

Or, [NV1, NV2, NV3, NV4] =[|A- BV|, |B- BV|, |C- BV|, |D- BV|] (2)
The values obtained, from the subtraction operations, are absolute values and are represented by NV1, NV2, NV3 and
NV4. The mode of this new data set (NDS) is again calculated to find if the new absolute values is equal to another or
simply occurs most frequently. This is done in order to find the one neighbor whose value is nearly equal to the value
yielded by the bilinear interpolator. This is done as per the boundary conditions set in Table 1.

Table 1: Boundary conditions.

Boundary Conditions Description
Minimum mode value In this case, the first value of the NDS is taken, as the mode (0.1) has minimum
value in date set. The selection of the first value can be achieved based on the
Example: subscripted indexing theory. This value is then subtracted from its bilinear value
NDS=[0.1, 0.1, 0.1, 0.7] and the absolute of this difference is assigned to the empty pixel.
Maximum mode value In this case, the mode (0.7) has maximum value in the data set. In this case, the
indices for the elements less or equal to the mode is calculated and arranged in
the ascending order. The first minimum value is then selected. The neighbor
Example: corresponding to this minimum value is then identified and the absolute
NDS=[0.1, 0.7, 0.7, 0.7] difference between this value and bilinear value is computed and assigned to the
empty location.
Zero mode In this case, there is no mode as the two data/elements of a set repeat the same
number of times. In this case, the first minimum value is considered and the pixel
Example: neighbor corresponding to that minimum number is identified. This can be
NDS=[0.1, 0.1, 0.7, 0.7] achieved by subtracting the minimum difference from the bilinear value, because

Volume 2, Issue 5, May 2013 Page 307

International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: [email protected], [email protected]
Volume 2, Issue 5, May 2013 ISSN 2319 - 4847

the minimum value is equal to the neighbor value minus the bilinear value. This
value is then assigned to the empty location.

Figure 3: The proposed zooming method.

Thus, the overall aim of finding this value (E) in this way, is to minimize the underestimation or overestimation of
some parts of image texture after undergoing the interpolation process because of the problems caused by the lowpass
filtering processes involved in many linear interpolators, bilinear in particular.

III. RESULTS AND DISCUSSIONS

The algorithm proposed above is tested for many pictures at different zoom scales. The quality of the algorithm is
evaluated in terms of Peak Signal to Noise Ratio against the conventional nearest and bilinear interpolation algorithms
using two grayscale images shown in Figure 3. The original image is downsampled to reciprocal of zoom scale. The
downsampled image is then zoomed again to its original size to compute the PSNR. The obtained results are tabulated
in Table 2. A higher peak signal to noise ratio would normally indicate the higher quality of the output image.

Original Image Downsampled Image Nearest Neighbor

Bilinear Proposed
Figure 4: Results for image Lena (size 144 x 144 pixels).

Volume 2, Issue 5, May 2013 Page 308

International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: [email protected], [email protected]
Volume 2, Issue 5, May 2013 ISSN 2319 - 4847

Original Image Downsampled Image Nearest Neighbor

Bilinear Proposed
Figure 5: Results for image Starfish (size 154 x 154 pixels).

Table 2: Computed PSNR for the zoomed images.

Picture Zoom Scale PSNR(dB)
Nearest Neighbor Bilinear Proposed
Lena 2 32 32.8 35
Starfish 2 35.2 35.8 37

It can be seen from Figures 4 & 5 and Table 2 that the proposed algorithm produces a better quality picture compared
to nearest neighbor and bilinear interpolation approaches. A high PSNR (Table 2) is obtained for images zoomed using
the proposed algorithm, compared to the nearest neighbor and bilinear interpolation approaches.

IV. CONCLUSIONS
This paper presented an adaptive image zooming technique based on the nearest neighbor and bilinear interpolation.
The selection between the two interpolation techniques depends upon the mode of the nearest pixels. Depending on the
interpolation ratio selected or set (i.e. depending on the final size desired/targeted), the interpolation algorithms,
mentioned here, gave different PSNR as well as visual quality. It is observed that this algorithm gives comparatively
higher values of PSNR and better visual appearance. It has been also observed that the images zoomed using the
nearest neighbor approach shows a texture with edge jaggedness while the second one (i.e. image interpolated using the
bilinear algorithm) shows comparatively soft but blurred texture. However, the proposed algorithm produces a better
quality smoother and sharper texture.

REFERENCES
[1.] M. Zamani, “An Applied Two-Dimensional B-spline Model for Interpolation of Data”, International Journal of
Advanced Research in Engineering and Technology (IJARET), Volume 3, Number 2, July-December (2012).
[2.] M. Nagaraju Naik and P. Rajesh Kumar, “Spectral Approach To Image Projection With Cubic B-Spline
Interpolation”, International journal of Electronics and Communication Engineering &Technology (IJECET),
Volume3, Issue3, pp. 153 - 161, 2012.
[3.] Zamani, M. “A simple 2-D interpolation model for analysis of nonlinear data”, Natural Science, Vol. 2, No. 6, P.
641-646, June 2010.
[4.] T. Acharya & P.S. Tsai, “Computational Foundations of Image Interpolation Algorithms”, ACM Ubiquity Vol. 8,
2007.
[5.] Z. Min, W. Jiechao, LI Zhiwei and LI Yonghua , “An Adaptive Image Zooming Method with Edge Enhancement”,
3rd International Conference on Advanced Computer Theory and Engineering (ICACTE), pp-608-611,
2010.

Volume 2, Issue 5, May 2013 Page 309

 International Journal of Application or Innovation in Engineering & Management (IJAIEM)
Web Site: www.ijaiem.org Email: [email protected], [email protected]
Volume 2, Issue 5, May 2013 ISSN 2319 - 4847

[6.] Zhou Dengwen and Shen Xiaoliu, “An Effective Color Image Interpolation Algorithm”, 4th International
Congress on Image and Signal Processing, pp-984-988, 2011.
[7.] J. Tao, Lu Wenhu, D. Shouyin and Han Mingqin, “Study of Image Magnification Based on Newton Interpolation”,
3rd International Congress on Image and Signal Processing, pp-1248-1252, 2010.
[8.] M. F. Fahmy, T. K. Abdel Hameed and G. F. Fahmy, “A Fast B-spline Based Algorithm for Image zooming and
Compression”, 24th National Radio Science Conference, Egypt, 2007.
[9.] M. Faghih and H. Pourghassem, “A Novel Interpolation Method Using Radon Transform”, IEEE 7th International
Colloquium on Signal Processing and its Applications, pp-277-282, 2011.
[10.] T. Acharya and Ping-Sing Tsai, “Computational Foundations of Image Interpolation Algorithms”, vol.8, 2007.
[11.] G.Ramponi, “Warped Distance for Space-Variant Linear Image Interpolation”, IEEE Transactions on Image
Processing, vol.8 n.5, May 1999, pp. 629–639.

Volume 2, Issue 5, May 2013 Page 310

View publication stats