Docstoc

SPATIAL DOMAIN IMAGE ENHANCEMENT USING

Document Sample
SPATIAL DOMAIN IMAGE ENHANCEMENT USING Powered By Docstoc
					              Journal of Electronics and JOURNAL OF ELECTRONICS AND
InternationalINTERNATIONAL Communication Engineering & Technology (IJECET), ISSN 0976 –
6464(Print), ISSN 0976 – 6472(Online) Volume 3, Issue 2, July-September (2012), © IAEME
    COMMUNICATION ENGINEERING & TECHNOLOGY (IJECET)

ISSN 0976 – 6464(Print)
ISSN 0976 – 6472(Online)
Volume 3, Issue 2, July- September (2012), pp. 209-216
                                                                          IJECET
© IAEME: www.iaeme.com/ijecet.html
Journal Impact Factor (2012): 3.5930 (Calculated by GISI)               ©IAEME
www.jifactor.com




             SPATIAL DOMAIN IMAGE ENHANCEMENT USING
                   PARAMETERIZED HYBRID MODEL
                               1
                                   I.Suneetha and 2Dr.T.Venkateswarlu
               1
                 Associate Professor,ECE Department,AITS,Tirupati,INDIA Pin-517520.
    2
      Professor,ECE Department,S.V.University College of Engineering,Tirupati,INDIA Pin-517501.
                         1
                           iralasuneetha.aits@gmail.com, 2 warlu57@gmail.com

Abstract
Images are very powerful tools to provide information to the viewers in every field i.e. medical
images for doctors, forensic images for police investigation, text images for readers etc. In the
process of image acquisition, contrast of an image becomes poor because of lighting, weather,
distance, or equipment used for image capture. Noise corrupts the images during sensing with
malfunctioning cameras, storing in faulty memory locations or sending through a noisy channel.
Sometimes quality of the image may be corrupted by poor contrast and unwanted noise. This
paper proposes a method for image enhancement through contrast improvement and noise
suppression using a Parameterized Hybrid Model in spatial domain. The proposed method
provides good results subjectively as well as objectively for both gray scale and true color
images. The proposed method is better, faster, and also useful for interactive image processing
applications as it provides various enhancement images for an image.

Key Words-Parameterized Gradient Intercept (PGI), Parameterized Adaptive Recursive (PAR), Mean
Square Error (MSE), Peak Signal to Noise Ratio (PSNR), and Digital Image Processing(DIP).


             I.     Introduction                     them using MATLAB [1]. These techniques
                                                     have been extended successfully to true
       Image enhancement improves digital            color images [2]. Image enhancement
image quality without knowing the source of          through noise suppression can be done using
degradation and provides visually acceptable         a Nonlinear Parameterized Adaptive
images for human viewers and/or automated            Recursive (PAR) model [3]. Image
image processing techniques. We reviewed             enhancement through contrast improvement
enhancement techniques for gray scale                can be done by using a Linear Parameterized
images in spatial domain and implemented             Gradient Intercept (PGI) model [4]. Linear
                                               209
International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 –
6464(Print), ISSN 0976 – 6472(Online) Volume 3, Issue 2, July-September (2012), © IAEME

and nonlinear models work well when an
image is corrupted by either poor contrast or
unwanted noise, but fails when corrupted by
both. This paper proposes a method for
image enhancement through contrast
improvement and noise suppression using a
Parameterized Hybrid model in spatial
domain. The type of noise considered is salt
and pepper noise. Sections II and III cover
related work done about linear PGI model
and nonlinear PAR model.

           II.    Linear PGI Model

       Relation between the input image
and output image in a linear PGI model is
                               0≤x<
     g x, y = G × f x, y + I
                               0≤y<
where G is Gradient and I is Interception of
the transformation. G and I values can be
zero, positive, or negative. When G and/or I
values are varied for improving the image
contrast, above transformation becomes
simple linear or nonlinear but not
exponential or logarithmic as in traditional
point processing methods and does not
require PDF calculations as in histogram                            (a)                  (b)
processing operations.                                  Fig. 1: (a) Man image, darken image, THE image,
                                                        AHE image and PGI image (b) Their Histograms
        PGI model works well for a gray
scale image and results are much more                          Table 1: MSE and tc for Man image
pronounced for true color image by                             MSE                          tc
preserving maximum color details. Results              THE     AHE      PGI       THE     AHE       PGI
indicate that mean square error (MSE) and             0.0124   0.0972   2.5e-9   0.0057   0.0334   0.0003
Computational time (tc) of PGI method is
smaller when compared to Traditional                            III.    Nonlinear PAR Model
Histogram Equalization (THE) and Adaptive
Histogram Equalization (AHE) methods.                    The relation between input image and
                                                      output image for nonlinear PAR model is

                                                               g(x,y) = imed[fn(x,y)]

                                                      where imed means Intentional median filter
                                                      that performs filtering to noisy pixels
                                                      intentionally. Let A be the window size that
                                                      is adaptive and R be the Recursive order. A

                                                210
International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 –
6464(Print), ISSN 0976 – 6472(Online) Volume 3, Issue 2, July-September (2012), © IAEME

and/or R can be varied. Results indicate that             combinable to enhance an image to that is
PAR method has small tc and high PSNR                     corrupted by both poor contrast and
when compared to TMF, RMF, and AMF.                       unwanted noise. As we are combining a
                                                          linear and nonlinear model, the resultant
       Table 2: PSNR and tc for Man image                 model can be named as Parameterized
                   PSNR(dB)
                                                          Hybrid Model (PHM) in which G and/or I,
 Original   TMF RMF            AMF        PAR
  63.29     74.25 73.87        71.54      81.39           A and/or R are varied.
                    tc( sec)                                      The proposed PHM has smallest
  TMF          RMF          AMF          PAR              MSE and highest PSNR at low
0.439644     0.633756     0.370175      0.182324          computational cost in spatial domain. The
                                                          following are the steps involved in PHM
                                                          algorithm simulation.

                                                          Gray scale image:

                                                            1.     Consider a good contrast and noiseless
                                                                  image i(x,y).
                                                            2.    Get poor contrast image fc(x,y) by
                                                                  amplitude scaling of i(x,y)
       (a)               (b)           (c)
                                                            3.    fn(x, y) is a noisy image of fc(x,y).
                                                            4.    Select appropriate values of A and R.
                                                            5.    Ensquare noisy image with (A-1)/2
                                                                   zeros to get padded image fp(x,y)
                                                            6.    gp(x,y) is imed filter of fp(x,y) .
                                                            7.     If gp(x,y) is noisy, vary A and/or R.
                                                            8.    If A varies, go to 4th step otherwise
                                                                   go to 5th step.
                                                            9.    Remove the ensquared zeros in gp(x,y) to
                                                                  get denoisy image fdn(x,y).
                                                            10.   Select appropriate values of G and I.
                                                            11.   Multiply fdn(x,y) by G and add I to get
                                                                  g(x,y)
                                                            12.   Observe the enhanced image g(x,y).
        (d)               (e)             (f)               13.   If g(x,y) is not good in contrast, then
    Fig. 2: (a) Man image (b) noisy image (c) TMF                 change G and/or I, go to11th step.
        (d) RMF (e) AMF, and (f) PAR images
                                                          True Color image:
             IV.    Proposed Method                          1. Consider a good contrast and noiseless
                                                                image i(x,y).
       PGI model improves contrast with                      2. Get poor contrast image fc(x,y) by
smallest mean square error and low                              amplitude scaling of i(x,y)
computational time. PAR model suppresses                     3. fn(x, y) is a noisy image of fc(x,y).
noise with highest PSNR and low
                                                             4. Select appropriate values of A and R.
computational time. Hence linear PGI model
                                                             5. Extract r,g,b components from fn(x,y)
and nonlinear PAR model can be
                                                    211
International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 –
6464(Print), ISSN 0976 – 6472(Online) Volume 3, Issue 2, July-September (2012), © IAEME

   6. Select appropriate values of A and R.                18. If g(x,y) is not good in contrast, then
   7. Ensquare noisy rgb images of fn (x,y)                    change G and/or I, go to15 th step.
       with (A-1)/2 zeros to get padded images                            V. Results
       rpgpbp.
   8. Perform imed filter to rpgpbp separately.
                                                                The PHM performance can be
   9. Get color image gp(x,y) from filtered             verified by not only by visual inspection of
       rpgpbp.                                          the resultant images but also by evaluating
   10. If gp(x,y) is noisy, vary A and/or R .           the mean square error and Peak Signal to
   11. If A varies go to 7th step otherwise             Noise Ratio in decibels (PSNR) [5-7]. The
       go to 8th step.                                  subjective results and objective results are
   12. Remove the ensqured zeros in gp(x,y)             shown in the following figures and tables.
       to get denoisy image fdn(x,y).
                                                                                       −1
   13. Extract Y from fdn(x,y) using RGB to                               = 20
       YIQ conversion to get l(x,y).
   14. Select appropriate values of G and I.
   15. Multiply l(x,y) by G and add I to get                          1
       f(x,y).                                                    =               ,    −    ,
   16. Get enhanced image g(x,y) from f(x,y)
       using YIQ to RGB conversion.
   17. Observe the enhanced image g(x,y).




                                                  212
International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 –
6464(Print), ISSN 0976 – 6472(Online) Volume 3, Issue 2, July-September (2012), © IAEME




                                                213
International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 –
6464(Print), ISSN 0976 – 6472(Online) Volume 3, Issue 2, July-September (2012), © IAEME




              (a)                               (b)                     (a)                  (b)
   Fig. 3: (a) Darken and noisy gray scale images           Fig. 4: (a) Darken and noisy true color images
           (b) Parameterized Hybrid Model images                    (b) Parameterized Hybrid Model images


                                                      214
International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 –
6464(Print), ISSN 0976 – 6472(Online) Volume 3, Issue 2, July-September (2012), © IAEME




           (a)                (b)
      Fig. 5: (a) Brighten and noisy gray scale                         (a)                (b)
                        images                          Fig. 6: (a) Brighten and noisy true color images
              (b) Parameterized Hybrid Model                   (b) Parameterized Hybrid Model images.
                        images

                                                  215
International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 –
6464(Print), ISSN 0976 – 6472(Online) Volume 3, Issue 2, July-September (2012), © IAEME


        Visual inspection of subjective               when an image is corrupted differently in
results indicates that, the Parametric Hybrid         various regions.
Model works very well by enhancing gray
scale and true color images that are                  REFERENCES
corrupted by decreased contrast and                    [1]   Ms.     I.Suneetha    and      Dr.T.Venkateswarlu,
                                                             “Enhancement Techniques for Gray scale Images in
unwanted noise. Visual inspection of
                                                             Spatial Domain”, International Journal of Emerging
objective results shows that, MSE was                        Technology and Advanced Engineering, website:
decreased and PSNR was increased for the                     www.ijetae.com(ISSN 2250-2459) Volume 2, Issue 4,
gray scale images and also for the R, G, and                 April 2012, pp.13-20.
B components of true color images. The
limitation in the proposed model is small              [2]   Ms.     I.Suneetha       and     Dr.T.Venkateswarlu,
                                                             “Enhancement Techniques for True Color Images in
decrement in mse and small improvement in                    Spatial Domain”, International Journal of Computer
PSNR for enhancing gray scale and true                       Science    &      Technology      (IJCST),  Website:
color images that are corrupted by increased                 www.ijcst.com(ISSN 0976-8491) Volume 3, Issue 2,
contrast and unwanted noise. The reason is,                  Version 5, April - June 2012, pp. 814-820.
while increasing contrast for getting
                                                       [3]   Ms. I.Suneetha and Dr.T.Venkateswarlu, “Image
simulation results some pixel values reach
                                                             Enhancement Through Noise Suppression Using
maximum value which are then treated as                      Nonlinear Parameterized Adaptive Recursive Model”,
salt during denoising. Therefore resultant                   International Journal of Engineering Research and
images are not having very good contrast.                    Applications (IJERA), Website: www.ijera.com (ISSN
This problem can be overcome by slight                       2248-9622), Volume 2, Issue 4, July-August 2012, pp.
change in PHM algorithm.                                     1129-1136.
               VI. Conclusions                         [4]   Ms. I.Suneetha and Dr.T.Venkateswarlu, “Image
        Spatial Domain Image enhancement                     Enhancement Through Contrast Improvement Using
using Parameterized Hybrid Model has been                    Parameterized Gradient Intercept Model”, ARPN
successfully implemented using MATLAB.                       Journal of Engineering and Applied Sciences (ARPN-
This paper considers gray scale and true                     JEAS), Website:www.arpnjournals.com           (ISSN
color images from different fields. Choice of                1819-6608), Volume 7, No. 8, August 2012.

A and R depends on noise intensity where as            [5]   J Rafael C Gonzalez, Richard E. Woods, and Steven L.
choice of G and I depend on amount of poor                   Eddins, Digital Image Processing Using MATLAB®
contrast. As proposed algorithm is a faster                  (Second Edition, Gates mark Publishing, 2009).
and better, PHM can be used as a tool for
                                                       [6]   J. Y. im, L. S. Kim, S. H Hwang, “An advanced
Photo editing software like Photoshop or
                                                             Contrast Enhancement Using Partially Overlapped
any existing image processing software by                    Sub–Block     Histogram       Equalization”, IEEE
attaching two sliding bars for A and R that                  Transactions on Circuits and Systems for Videc
suppresses noise and two sliding bars for G                  Technology, Vol. 11, No. 4, pp.475-484,2001.
and I that improves contrast. The PHM
                                                       [7]   Prof. A. Senthilrajan, Dr. E. Ramaraj, “High Density
model can be used for suppressing high
                                                             Impulse Noise Removal in Color Images Using Region
level salt and pepper noise or other types of                of Interest Median Controlled Adaptive Recursive
noises with slight changes in algorithm.                     Weighted Median Filter”, Proceedings of the
Future scope will be the development of                      International MultiConferenceof Engineers and
local parameterized models for image                         Computer Scientists (IMECS), Vol. II, March 17-19,
enhancement in Region Of Interest (ROI)                      2010, Hong Kong.



                                                216

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:13
posted:11/20/2012
language:
pages:8