Scholarly article on topic 'Performance Evaluation of Fuzzy and Histogram Based Color Image Enhancement'

Performance Evaluation of Fuzzy and Histogram Based Color Image Enhancement Academic research paper on "Materials engineering"

CC BY-NC-ND
0
0
Share paper
Academic journal
Procedia Computer Science
OECD Field of science
Keywords
{Fuzzy / "Contrast Enhancement" / Histogram / "Color Images"}

Abstract of research paper on Materials engineering, author of scientific article — Taranbir Kaur, Ravneet Kaur Sidhu

Abstract In computer vision applications, image enhancement plays an important role. It has been analyzed that the majority of the existing enhancement techniques are in relation to the transform domain methods, which can introduce the color artifacts and also may reduce steadily the intensity of the input image. To overcome this dilemma, a fuzzy based algorithm has been used. This approach could have the ability to boost the contrast in digital images in an efficient manner by utilizing the histogram based fuzzy image enhancement algorithm. The overall objective of this paper is to evaluate the effectiveness of histogram and fuzzy based image en-hancement for various kinds of images like underwater, remote sensing, medical etc. The fuzzy and histogram based enhancement has been designed and implemented in MATLAB using image processing toolbox. The result has shown the effectiveness of the fuzzy based enhancement over the existing techniques

Academic research paper on topic "Performance Evaluation of Fuzzy and Histogram Based Color Image Enhancement"

(8)

CrossMark

Available online at www.sciencedirect.com

ScienceDirect

Procedia Computer Science 58 (2015) 470 - 477

Second International Symposium on Computer Vision and the Internet (VisionNet'15)

Performance Evaluation of Fuzzy and Histogram Based Color Image

Enhancement

Taranbir Kaura, Ravneet Kaur Sidhub

a CT Institute of Technology and Research, Jalandhar, India b CT Institute of Technology and Research, Jalandhar, India

Abstract

In computer vision applications, image enhancement plays an important role. It has been analyzed that the majority of the existing enhancement techniques are in relation to the transform domain methods, which can introduce the color artifacts and also may reduce steadily the intensity of the input image. To overcome this dilemma, a fuzzy based algorithm has been used. This approach could have the ability to boost the contrast in digital images in an efficient manner by utilizing the histogram based fuzzy image enhancement algorithm. The overall objective of this paper is to evaluate the effectiveness of histogram and fuzzy based image enhancement for various kinds of images like underwater, remote sensing, medical etc. The fuzzy and histogram based enhancement has been designed and implemented in MATLAB using image processing toolbox. The result has shown the effectiveness of the fuzzy based enhancement over the existing techniques

© 2015PublishedbyElsevierB.V.Thisisanopenaccess article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of organizing committee of the Second International Symposium on Computer Vision and the Internet (VisionNet'15)

Keywords: Fuzzy; Contrast Enhancement; Histogram; Color Images

1. Introduction

In a computer vision system, contrast image enhancement technique is used in order to improve the visual quality of image and to analyze the specific image features for future processing, which are hardly detectable by human vision1. The main purpose of image enhancement is to process a given image so that the outcome is more appropriate than the original image for a definite use, such as segmentation and identification of objects. The techniques of contrast development perform very well with images having a uniform spatial distribution of gray values, but problem occurs in case of non uniform distribution of brightness. To overcome the problems with low contrast images, an easy and efficient fuzzy based automatic contrast enhancement method is used here, which helps to improve the visual quality of image as well as aids in the easy extraction of the spatial features present in the image. The used technique is

* Corresponding author. Tel.: +91-9501722979 E-mail address: virk.taranbir@yahoo.coma, ravneet89.sidhu@gmaiï.comè

1877-0509 © 2015 Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of organizing committee of the Second International Symposium on Computer Vision and the Internet (VisionNet'15) doi: 10.1016/j.procs.2015.08.009

centred on two variables: the average intensity value of the image M and, contrast intensification parameter K. The given approach uses HSV color space in which only V part is stretched preserving the chromatic data (H and S).

Nomenclature

M intensity value

K intensification parameter

x current intensity value

xe improved intensiy value

h(x) histogram

a1 class

a2 class

v1 membership function for a1 class

v2 membership function for a2 class

Ve improved component

2. Literature Survey

Choi YS et al.2 have proposed a robust approach based on fuzzy logic. It handles the conflicting goals of the image enhancement by removing the impulse noise, smoothing the non impulse noise, and by edge enhancement. To perform the above tasks, three different filters are used. Hanmandlu et al.3 have introduced a global contrast intensification operator for enhancement of color images with three intensification parameters. The proposed technique helps to improve the visual quality of under exposed images and also handles over exposed images by using maximum intensity as the fourth parameter. Agaian S. et al.4 have discussed three techniques for image enhancement which include logarithmic transform histogram moving, logarithmic transform histogram matching, and logarithmic transform histogram shaping using Gaussian distributions. For contrast enhancement a quantitative measure is also defined which is based on human visual system. Karen et al.5 introduced two enhancement algorithms: edge-preserving contrast enhancement to preserve edge details and, a multi histogram equalization method to enhance the contrast of images allowing a fast and efficient correction of non uniform illumination. Rekha L. et al.6 have introduced an Automatic Selective Grey-Level grouping method to transform the skewed histogram into uniform histogram. This technique does not require any user interface. Results obtained by this method are better than histogram based gray level grouping and fuzzy logic method. Jin et al.7 have proposed a histogram and wavelet based industrial X-ray image enhancement algorithm. The proposed technique is capable to deal with low contrast and poor quality details. This algorithm can recover the global image contrast successfully and defeat the observable artifacts of X-ray image. Fang et al.8 have discussed a method to improve the information of the image, by using enhancement method. Region-based enhancement is performed and the regions which do not need enhancement get dropped. Yaping et al.9 have proposed a step wise method which includes: image pre-processing, image recognition and enhancement that provides a procedural guide for complex image recognition. Weizhen et al.10 have given a method to gain the vision quality of the image. Retinex enhances the image by processing reflection R and incident light. It has been described that the reflection of the light varies from black to white for image and from black to colorless for transparent volume. Retinex algorithm improves the local and global enhancement for medical images Zhang et al.11 have proposed homomorphism enhancement method to remove the irregular explanation in frequency domain algorithm. Hybrid algorithm is also offered to enhance the image and to enhance the image details in frequency domain the Gauss filter processing is used. Tianhe et al.12 has proposed a multifractal theory of infrared images to resolve the problem of visibility and blurry edges. It also extracts the multifractal characteristics of edges in infrared images. The singularity characteristics of each pixel in the image are calculated and for pixel classification, Human Visual System is used. Verma et al.13 proposed a gray-scale based image enhancement method that increases their natural contrast by using some constant parameters. On the basis of light and dark edges, the local contrast gets enhanced. To describe digital image, local mean and local standard deviation of full image with minimum and maximum value are used. Sun et al.14 have proposed a new optical transfer function-based micro image enhancement algorithm. The point spread function is gained from uneven

brightness in optical system and to obtain the optical transfer function the optical property based high pass filter is constructed. The optical transfer function-based micro image enhancement algorithm provides a better micro image enhancement effects. Imtiaz et al.15 have given a method to enhance images at gray level and color reproduction. Fuji intelligent Color Enhancement technique is used to convert the RGB endoscopic image into two-dimensional gray scale spectral. The image with optimum entropy is chosen as enhanced image. Raju et al.16 proposed a new fuzzy logic and histogram based algorithm to enhance the low contrast color images. In this approach, RGB image is converted into HSV and under the control of two parameters M and K, V Component is stretched, where M is considered as average intensity and K is contrast intensification.

3. Contrast Enhancement Techniques

Conventional processes for contrast enhancement include gray-level transformation based practices (viz., logarithm transformation, power-law transformation, piecewise-linear transformation, etc.) and histogram based running methods (viz., histogram equalization (HE), histogram specification, etc.

3.1. Histogram equalization

The most commonly used approach is histogram equalization based on the assumption that the uniformly distributed gray scale histogram has the best visible contrast. Different sophisticated histogram based enhancement methods include Bi-Histogram Equalization (BHE), Block-Overlapped Histogram Equalization, Multi-Scale Adaptive Histogram Equalization1, Form Keeping Local Histogram. These techniques are used to modify image intensities in order to enhance contrast. Let 'f' be a 'Mr' x 'Mc' confirmed image, having intensities ranging from 0 to [L -1]. 'L' is the amount of possible depth values; frequently 256. The small histogram distributions represent low contrast images and wide histogram distributions represent high contrast images. This technique is particularly helpful in images having large regions of related tone such as an image with a very light background and dark foreground. Histogram equalization can depict hidden details in an image by stretching out the contrast of local regions and hence making the differences in the regions more observable.

3.2. Adaptive Histogram Equalization

It is a computer image processing method used to enhance the image contrast. The technique is different from standard histogram equalization in the context that adaptive histogram 7 method computes different histograms corresponding to a definite part of the image, and employs them to reconstruct the brightness values of the image. It is thus ideal for increasing the local contrast of an image and bringing out more detail. However, AHE tends to over amplify noise in fairly homogeneous regions of an image. A plan of adaptive histogram equalization called contrast limited adaptive histogram equalization (CLAHE) prevents that by restraining the amplification.

3.3. Fuzzy Enhancement

Gray tone is modified with the help of membership function and a fuzzy set is obtained. Here the image is considered as a range of fuzzy singletons having a membership value. The value represents the degree of image property in the range. And intensification operator helps to modify the membership functions. Different membership values were used by different experts in order to improve the quality and sleep of contrast enhancement based on the fuzzy logic method. Recently, brand new intensification operator called NINT was proposed3, which works as sigmoid function for the change of the Gaussian form of membership. The method is based on the optimization of entropy by a parameter mixed up in intensification driver which will be suitable for gray level images.

4. Methodology

The given fuzzy enhancement16 technique is intended only for improving the color images with low contrast and low brightness. The method uses HSV color area where only the V element is expanded by protecting the chromatic

details such as Hue (H) and Saturation (S). Extension of V element is conducted under the control of the enhancement factors M and K. By stretching, the current intensity value (x) will be converted to improved intensity value (xe). According to the given method, a P Q dimensioned RGB image is converted into HSV and its histogram h(x) is evaluated, where x e V. h(x) represents the number of pixels in the image with intensity value x.

Fig. 1. Flowchart of the methodology

Here two intensification parameters M and K are used, which controls the level at which the intensity value x has to be increased. The control parameter M, an average intensity value of the image can be measured from the histogram equation as follows:

M = Wl <1>

Exh(x)

The parameter M separates the histogram h(x) into two classes: a1 and a2. The pixels having value in the range [0, M-1] lie in a1 class and pixels having values in the range [M, 255] lie in a2 class. To perform stretching, V component uses two fuzzy membership values: pv1 for class a1 and pv2 for class a2. The stretching intensity is decided by enhancement parameter K for calculating the improved intensity values xe for class a1 and a2. A stretching point is

decided by parameter K to which the intensity value x should be expanded in accordance with the account principles Hv1 and pv2.The value for K can be calculated empirically according to what level the extension is needed. From the trial research, the value of K is set to 128, which gives better outcomes for color images with low contrast and low brightness. For class a1 the value of fuzzy membership function pv1 is based on the factor that how far intensity value x is from parameter M. Fuzzy rule for a1 is can be defined as: The stretching intensity should be SMALL if the distinction between x and M is LARGE.

According to the above rule, the pixel value nearer to M will be stretched more whereas values away from M will be stretched less. Other pixels with value in between will be stretched proportionately. The following statistical expression can be used to implement the above fuzzy rule:

( ) 1 - (M - x)

P =-M--(2)

Where x e a1, after calculating the membership value of x. The following equation is used to compute the contrast enhanced value xe for class a1.

xe = x + pv1(x)K (3)

Further to get the improved value xe, pv1(x) chooses the quantity of extending parameter K to be added to x. For class a2, fuzzy membership value pv2(x) is based on the factor that how far the intensity value x is from the extreme value F (for 8 bit image F=255). The fuzzy rule for class a2 depends upon the difference between x and F. If the difference is LARGE, the stretching intensity should also be LARGE. According to above rule, pixels whose value is closer to F will be less extended and the pixels whose value is away from F will be more extended. Pixels value in between will be extended proportionately. Rule can be defined by following equation:

^(x) = F-M (4)

Where x e a2, after calculating the membership value of x. The following equation is used to compute the contrast enhanced xe for class a2:

xe = (xpv2(x)) + (F - Pv2(x)K) (5)

pv2(x) Chooses the quantity of extending parameter K and the intensity value x that has to be used to get the improved value xe. The old x values of the V component will be substituted with the improved xe values, expanding the V component into contrast and brightness improved component Ve. This improved achromatic detailed component Ve along with the preserved chromatic details (Hue and Vividness components) forms the improved picture HS Ve This HS Ve image is then transformed into improved RGBe image. The flowchart of the given technique is shown in Fig.1.

5. Experiments ans Results

In this section, we demonstrate the performance of various contrast enhancement techniques in terms of Mean Square Error (MSE) and Peak Signal to Noise Ratio (PSNR). The implementation has been done in MATLAB 7.12 under the Windows 8.0 on a PC with CORE i5 CPU and 4 GB RAM. Experiments have been performed on 10 images with different contrast and brightness, and results of one of the test image have been shown by Fig.2 and 3.

5.1. Contrast image enhancement using Histogram equalization, Adaptive histogram and Fuzzy enhancement:

The results show the performance of various enhancement algorithms on low contrast and low bright images. Moreover various qualitative performance measures has been used to prove the superior performance of the fuzzy method over convention and advanced methods. Fig.2. shows the images before and after the contrast enhancement.

Fig.3. shows the histograms of original low contrast, low brightness and contrast enhanced images. It has been analyzed that the result shown by adaptive histogram and fuzzy enhanced technique is better as compared to the input image and histogram equalized image.

Fig. 2. (a) Input Image;(b) histogram equalized Image;(c) adaptive histogram equalized Image;(d) fuzzy enhanced image. :--■-----

fa ^ 7

Fig. 3. Shows the histograms of above image (a) Input Image;(b) histogram equalized Image;(c) adaptive histogram equalized Image;(d) fuzzy enhanced image.

5.2. Performance Evaluation

• Mean Square Error:

. In image processing, mean square error17 is the most general measure for performance measurement of the existing method and the coded images. It is computed by using equation:

1 i=0 j=1

MSE = ^ Z Z(f a* j) - f j))2 (6)

In above equation R and C represent the number of rows and columns in the input images with index i and j respectively. f(i,j) represents the original image at location (i, j) and f (i,j) represents the degraded image at location (i,j). Table 1 show the values for Mean Square Error for various color images and Fig.4. shows the analysis of the mean square error of the various input images with different techniques. It is clear from the plot that MSE value obtained by fuzzy method is comparatively low than other equalization techniques.

• Peak Signal to Noise Ratio:

PSNR17 refers to the ratio between the maximum possible power of a signal and the power of corrupting noise that affects the quality of image. Higher value of PSNR indicates that the reconstruction is of higher quality. Where MAXi represents maximum peak signal value that exists in original image. Table 1 shows the values for Peak Signal to Noise Ratio for various color images. Fig.5. shows the analysis of the Peak Signal to Noise Ratio of the various input images with different techniques. It is clear from the plot that PSNR value obtained by fuzzy method is increased as compare to other equalization techniques. PSNR is defined as:

MAX MAXI

PSNR = 10. log1o(—) = 20. log1o(——1) = 20. log^(MAXi) - 10. log^MSE) (7)

Table 1. Performance evaluation after applying various techniques on different images.

Img 1 Img 2 Img 3 Img 4 Img 5 Img 6 Img 7 Img 8 Img 9 Img 10

HE 3710 822 1717 1870 143 3029 865 11796 344 9732

MSE AHE 2021 2370 403 1759 101 3472 1344 16790 803 6712

Fuzzy 21.9629 6.1211 4.4016 20.2192 8.9466 18.3027 36.8365 486.6866 12.2933 103.307

PSNR HE AHE Fuzzy 12.4371 15.0751 34.7139 18.9821 14.3833 40.2625 15.7831 22.0778 41.6947 15.4124 15.6781 35.0732 26.5774 28.0876 38.6094 13.3178 12.7250 35.5056 18.7606 16.8468 32.4680 7.4135 5.8803 21.2583 22.7652 19.0836 37.2341 8.2488 9.8623 27.9895

16000r 16000 ■ 14000 ■ 12000 ■ 10000 ■ 8000 -6000 ■ 4000 ■ 2000 ■

I Histogram equiliraticn ] Adaptive his lo g rain equiliszation I Fuzzy

1 2 3 4 5 S 7

images

I Histogram equiiization J Adaptive histogram equiliazation I Fuzzy

1 2 3 4 5 B 7 Images

Fig. 4. Mean Square Error Analysis

Fig.5. Peak Signal to Noise Ratio Analysis.

6. Conclusion and future Scope

The fuzzy based image enhancement approach has the ability to boost the contrast in digital images in efficient manner by utilizing the histogram based fuzzy image enhancement algorithm. This approach is computationally fast as compare to other techniques .In this paper, we have evaluated the effectiveness of histogram and fuzzy based image enhancement technique in terms of MSE and PSNR. The results have shown the effectiveness of the fuzzy based enhancement technique with improved visual quality of the image. In near future, we will introduce a modified approach having the capability of enhancing the contrast in digital images efficiently by using the modified edge-preserving-smoothing-hypothesis based adaptive k-fuzzy based enhancement algorithm. The value of parameter k will be evaluated automatically by using some optimization technique.

References

1. Gonzalez RC, Woods RE.Digital Image Processing. Prentice-Hall, 2nd ed. 2002.

2. YoungSik C., Krishnapuram R. "A robust approach to image enhancement based on fuzzy logic", IEEE Image Processing Transactions, vol.6, no.6, 1997, pp.808-825.

3. Hanmandlu M., jha D. "An optimal fuzzy System for color image enhancement", IEEE Image Processing Transactions, vol.15, no.10, 2006, pp.2956-2966.

4. Agaian SS., Silver B., PanettaKA. "Transform coefficient histogram-based image enhancement algorithms using contrast entropy", IEEE Image Processing Transactions, vol.16, no. 3, 2007, pp.741-758.

5. Panetta K A., Wharton E J., Agaian SS."Human visual system-based image enhancement and logarithmic contrast measure", IEEE Systems Man and Cybernetics Transactions, vol. 38, no. 1, 2008, pp.174-188.

6. Lakshmanan R., Nair MS., Wilscy M.,Tatavarti R. "Automatic contrast enhancement using selective gray-level grouping", International Journal on Signal and Imaging Systems Engineering, vol.3, no.2, 2010, pp.126-135.

7. Jin L., Wei W.Y., Wang L., Jie W., and Da Da W. "Industrial X-ray image enhancement algorithm based on adaptive histogram and wavelet", IEEE 6th International Forum on Strategic Technology (IFOST),vol. 2, 2011, pp.836-839.

8. Fang X., Liu J., Gu W., Tang Y."A method to improve the image enhancement result based on image fusion", in Proc. International Conference on Multimedia Technology (ICMT), 2011, pp. 55-58.

9. Yaping L., Jinfang Z., Fanjiang X., Xv S."The recognition and enhancement of traffic sign for the computer-generated image", in Proc. 4th International Conference on Digital Home (ICDH), 2012, pp.405-410.

10. Weizhen S., Fei L., Qinzhen Z. "The applications of improved retinex algorithm for x-ray medical image enhancement", in Proc. International Conference on Computer Science & Service System (CSSS), 2012, pp.1655-1658.

11. Zhang C., Ma L.N., Jing L N. "Mixed frequency domain and spatial of enhancement algorithm for infrared image", in Proc. 9th International Conference on Fuzzy Systems and Knowledge Discovery, 2012, pp.2706-2710.

12. Tianhe Y., Lijuan J., Dianlong Z., Shuang G."Enhancement of infrared image using multi-fractal based on human visual system", in Proc. International Conference on Measurement, Information and Control (MIC), 2012, pp.622-625.

13. Verma A., Goel S., Kumar N. "Gray level enhancement to emphasize less dynamic region within image using genetic algorithm", in Proc. 3rd International conference on Advance Computing (IACC), 2013, pp. 1171-1176.

14. Sun Y., Yin X. "Optical transfer function-based micro image enhancement algorithm", in Proc. International Conference on Communications Workshops (ICC), 2013, pp. 959-963.

15. Imtiaz M S., Khan T H.;Wahid K. "New color image enhancement method for endoscopic images", in Proc. International Conference on Advances in Electrical Engineering (ICAEE),2013, pp.263-266.

16. Raju G, Nair M.S. "A fast and efficient color image enhancement method based on fuzzy-logic and histogram", International Journal of Electronics and Communications (AEU), vol.68, no.3, 2014, pp. 237-243.

17. Lee C S., Kuo Y H. , Yu P T. "Weighted fuzzy mean filters for image processing", Fuzzy Sets and systems, Elsevier ,vol.89, no.2, 1997, pp. 157-180.