IJRIT International Journal of Research in Information Technology, Volume 2, Issue 6, June 2014, Pg: 308-315

International Journal of Research in Information Technology (IJRIT) www.ijrit.com

ISSN 2001-5569

Modified Contrast Compensation Algorithm for Noisy Frontlit and Back-lit images using Global Thresholding and Image Illumination Analysis Noel Thomas1, Shajin Prince2 1

2

PG Student, Electronics and Communication Department, Karunya University Coimbatore, Tamil Nadu, India [email protected]

Assistant Professor, Electronics and Communication Department, Karunya University Coimbatore, Tamil Nadu, India [email protected]

Abstract Conventional contrast enhancement methods were not successful in generating optimal enhanced images for those with front-lit and back-lit. Front-lit are the ones with a bright foreground and back-lit images have a brighter background. Conventional methods have two shortcomings. First, most of them need transformation functions and parameters that are specified manually. Second, spatial position segmentation method for splitting the image into its foreground and background is complex. This paper presents a modified contrast compensation technique that is parameter free. In this modified approach first the input image is unsharped to obtain a noise reduced image which is then segmented by global thresholding. Segmented image is then classified into corresponding image class and is transformed into YIQ color space. Image illumination analysis is done on the luminance of the transformed image to obtain the enhanced luminance. The luminance image is then transformed back to RGB color space to obtain the contrast enhanced image.

Keywords: Contrast enhancement, fuzzy classification, image class, back-lit, front-lit.

1. Introduction Contrast is the difference in luminance or color that makes an image distinguishable. Medical image processing, image processing, video processing and computer vision gives much importance to contrast enhancement as it is the important preprocessing step of all of these. Contrast enhancement enhances the overall quality of the image for better human visual perception. Current contrast enhancement algorithms can be grouped into global, local and hybrid algorithms. Global methods utilize the information of the entire image to enhance it. In local enhancement algorithms each pixel in an image is enhanced by using the information (luminance, saturation etc) of the pixels and its neighbors. Hybrid algorithm combines both the global and local enhancement approaches. In hybrid enhancement methods the entire image is first divided into different non-overly or overly regions and each of these regions is enhanced using the global methods. In the case of noisy front-lit and back-lit face images the conventional contrast enhancement methods fail to produce a satisfactory result. Conventional methods require the parameters of enhancement to be entered or selected manually. This implemented algorithm provides an automatic and parameter free approach for contrast enhancement of back-lit, front-lit and normal-lit images as well as noisy face images. Noel Thomas,IJRIT

308

IJRIT International Journal of Research in Information Technology, Volume 2, Issue 6, June 2014, Pg: 308-315

The input parameters are automatically found and the output parameters are generated. A transformation function (Piecewise Linear Transform) generates the output luminance corresponding to its respective input luminance in the enhanced image. C. C. Sun and S. J. Ruan proposed a contrast enhancement method based on histogram equalization. Dynamic Histogram Equalization (DHS) algorithm can be applied by simple hardware and processed in real-time system due to its simplicity. However, it tends to change the mean brightness of the image to the middle level of the gray-level range, which is not desirable in the case of images from consumer electronics products. D. Menotti, L. Najman and J. Facon proposed multihistogram equalization method for contrast enhancement and brightness preserving.

2. Modified Contrast Compensation Algorithm Noisy color face image

Unsharp Masking

Image Illumination Analysis

Global Thresholding

Image Compensation

Fuzzy logic classification

Color transformation (YIQ-RGB)

Color Transformation (RGB-YIQ)

Enhanced Image

Fig. 1 Modified Contrast Compensation Algorithm The input is a noisy color face image. The input can be back-lit, front-lit or normal-lit image. By unsharped masking the noise is effectively removed to a good extend. The noise free image is then separated into its corresponding background and foreground images and the image class is found by fuzzy classification. The RGB image is then transformed into YIQ color space. Image illumination analysis and image compensation is done on this color transformed image and is then transformed back to the RGB color space to obtain the optimal enhanced image.

2.1 Unsharp Masking The "unsharp" of the name derives from the fact that the technique uses a blurred, or "unsharp", positive image to create a mask of the original image. The unsharped mask is then combined with the negative image, creating an image that is less blurry than the original. The input noisy image is first median filtered to obtain a filtered or noise reduced image. This filtered image is then scaled by a factor k (less than 1). The scaled image is then subtracted from the original noisy image to obtain a mask of itself. The mask image thus obtained is then added with the original image to obtain a contrast enhanced and noise reduced image. The idea of unsharp masking filtering is the edge enhancement technique in which original image is filtered using a high pass filter and then scaling the filtered image using the constant value (k) where constant value is less than unity.

Noel Thomas,IJRIT

309

IJRIT International Journal of Research in Information Technology, Volume 2, Issue 6, June 2014, Pg: 308-315

Input image

Median Filter

Image Adder

Scaling(k)

Enhanced Image

Masking

Fig. 2 Unsharp Masking

2.1.1 Median Filtering High frequency signals can be enhanced by median filtering. Median filtering reduces the noise without disturbing the edges. Therefore median filter is an important filter in digital signal processing as it provides the edge enhancement. It is a type of non-linear noise filter where it prevents edges while reducing noise.

2.1.2 Scaling and Masking In this modified contrast enhancement algorithm the image is scaled by a constant factor which is less than one. The scaled image is then subtracted from the original image to obtain a mask of the image. This obtained mask is again added with the original image to give an enhanced image with reduced noise content.

2.2 Global Thresholding The segmentation of an image into its background and foreground is done by global thresholding. In global thresholding the segmentation is done based on a threshold value and hence its name. initial threshold is the average of all the image pixel values. The steps in global thresholding are the following, a) Initial estimate of threshold T b) Segmentation using T: G1, pixels brighter than T; G2, pixels darker than (or equal to) T; c) Computation of the average intensities m1 and m2 of G1 and G2 d) New threshold value:  


 =  (1) e) If |
−
 | > ∆
, back to step ‘b’ ,otherwise stop.

The value of ∆
is carefully chosen to obtain an optimal segmented image. Global thresholding segments the image into its central foreground (CF) and boundary background (BB) images.

2.3 Fuzzy logic classification Fuzzy logic is a convenient way to map an input space to an output space. This is the starting point for everything else, and the great emphasis here is on the word “convenient”. The luminance range of the boundary background (BB) and center foreground area (CF) is used to determine whether the image is back-lit or front-lit. The fuzzy inference method is used to determine the class to which the image belongs

Noel Thomas,IJRIT

310

IJRIT International Journal of Research in Information Technology, Volume 2, Issue 6, June 2014, Pg: 308-315

to. Image classes can be back-lit, front-lit or normal-lit. The luminance range is found by the following equations, If ( < 128)  =  ( +  ∗   ) ! =  ( −  ∗   ) else

 =  ( −  ∗   ) (2) ! =  ( +  ∗   ) where  ,   ,  ,   are the average luminance of the center foreground area, the standard deviation luminance of the center foreground area, the average luminance of the boundary background area, and the standard deviation luminance of the boundary background area, respectively.  and  are membership function which converts BB and CF into a fuzzy degree, respectively. k is set as 0.5. The fuzzy inference rules are characterized by a collection of fuzzy membership functions, logical operations, and IF-THEN rules. The two input variables BB and CF have three fuzzy sets, L (low illumination), M (medium illumination), and H (high illumination). The output variable IC has three fuzzy sets, BL (back-lit), FL (front-lit), and NL (normal-lit). Therefore, nine fuzzy interference rules are used here. Two representative rules are given below, 1. If BB is H and CF is L then IC is BL 2. If BB is L and CF is H then IC is FL Based on the centroid defuzzification method, the value of the IC corresponding to two given BB and CF values are obtained from the fuzzy inference. The inferred IC value represents the final estimated classification degree of the input image. That is, to determine the input image is backlit, normal-lit or frontlit images.

2.4 Color Transformation The RGB color image is transformed into YIQ color space. The three components of YIQ color space are Luma (Y), Hue (I), and Saturation (Q). The first component, luminance, represents grayscale information, and the last two components make up chrominance (color information). The transformation equations are given below, # 0.299 " $ & = "0.595716 % 0.211456 / 1 "0 & = "1  1

0.587 −0.2721 −0.522591

0.9563 −0.2721 −1.1070

/ 0.6250 −0.321263& "0 & 0.311135 

0.6210 # −0.6474& " $ & 1.7046 %

(3)

(4)

2.5 Image Illumination Analysis Each color image can be represented by Gaussian mixtures. Each single Gaussian distribution is expressed as one peak and two valleys at luminance component. From the transformed image luminance, hue and saturation are obtained. To analyze the illumination of the input image the luminance histogram, 23 (45 ) = 65 is first generated. The luminance histogram has its illumination levels from 0 to 255. 45 is the kth illumination level and 65 is the number of pixels in the image having illumination level 45 . The original luminance histogram consists of a large number of peaks and valleys thereby smoothing is essential. Luminance histogram is smoothened by Gaussian filtering. The Gaussian convolution is given by, A 789 :4, <= > = 29 (4) ∗ ?:4, <= > = @BA 29 () ?:4 − , <= >  (5) The average difference point x is defined by the first derivative of the smoothened histogram. A peak is defined as a positive to negative crossover in the first derivation of the smoothed histogram. Furthermore, a valley is defined as a negative to positive crossover. All peaks and valleys from the first derivation of the smoothed histogram are obtained. Noel Thomas,IJRIT

311

IJRIT International Journal of Research in Information Technology, Volume 2, Issue 6, June 2014, Pg: 308-315

2.6 Image Contrast Compensation Contrast compensation is based on piecewise linear transform (PLT). It is done by contrast stretching. Contrast stretching is the process that expands the range of intensity levels in an image so that it spans the full intensity range of the recording medium or displaying device. All the peaks and valleys obtained from the illumination analysis are the input parameters of the PLT. The transformation function is given by, 3 B3
5BC (4) = D DE (4 − 45BC ) + G5BC (6) FD BFDE

where H45 ,  = 0,1 … . J and HG5 ,  = 0,1 … J represents the input and output luminance respectively. The values of output parameters HGC , G , GK , … G5 J is given by, OD Pr(4) ∗ 255 (7) G5 = ∑FPQ

The Eq. (7) is used to enhance the image with normallit. In order to enhance back-lit or front-lit images, Eq. (7) must be adjusted. For the back-lit image, the foreground is located at dark illumination. The illumination value of black is smaller than 64. This set value 64 is according to the property of human vision perception stating that the brightness under 64 gray-level will be regarded as darkness. Thus, we adopt Eq. (7) to shift the darkest illumination to bright illumination. That is, the dark illumination can see more clearly. Herein, we introduce a BlackShift parameter to resolve this problem. The BlackShift parameter sets as BlackShift = 64 - MaxDarkPeak. The MaxDarkPeak is the max peak which locates at the dark illumination. Thus, the output parameters HGC , G , GK , … G5 J of back-lit images are defined as follows: OD G5 = RSTUV + ∑FPQ Pr(4) . 255 (8) For the front-lit image, the foreground is located at bright illumination. The background is located at dark illumination. In particular, the foreground is lighted by the sunlight. That is, the foreground is too bright to see. The bright luminance distributions need to be transformed from bright illumination spread to dark illumination. Thus, we introduce a WhiteShift parameter to resolve this problem. The WhiteShift parameter sets as WhiteShift = MaxBrightPeak - 223. This set value 223 is according to the property of human vision perception stating that the brightness above 223 gray-level will be regarded as brightness. Thus, the WhiteShift parameter is used to compress the brightness. Thus, the output parameters {y0, y1…yk} of the front-lit images are defined as follows: OD G5 = ∑FPQ Pr(4) . 255 − WUVXUV (9)

2.7 Color Transformation The compensated image is transformed back into RGB color space to obtain the enhanced image.

3.Experimental Results

Fig. 3a Input Noisy Back-lit Image

Noel Thomas,IJRIT

Fig. 4a Input Noisy Front-lit Image

312

IJRIT International Journal of Research in Information Technology, Volume 2, Issue 6, June 2014, Pg: 308-315

Fig. 3b Smoothened Luminance Histogram

Fig. 3c Transformation Function

Fig. 3d Enhanced Back-lit Image

Noel Thomas,IJRIT

Fig. 4b Smoothened Luminance Histogram

Fig. 4c Transformation Function

Fig. 4d Compensated Front-lit Image

313

IJRIT International Journal of Research in Information Technology, Volume 2, Issue 6, June 2014, Pg: 308-315

4.Performance Measurements Quantitative image enhancement is measured using PSNR and RMSE values. The higher value of PSNR gives the better performance and lower values of RMSE provides better enhancement. PSNR (Peak Signal to Noise ratio), which gives the quality of improvement, which is a maximization performance and is given by the following equation,

^^

YZ/ = 20R[?C\ ] c (10) _`ab RMSE (Root Mean Square Error) is the measure of error rate. It is the minimization function so the value should be minimum value. Root Mean Square Error (RMSE) is given by, /d7e = ]

C

`f

C/

fBC ^ ∑`BC FP\ ∑3P\ g$F3 − $F3 g c

where $F3 denotes input image and

^ $F3

(11)

represents enhanced image.

Performance Measure PSNR MSE

Table 1 Back-lit Image 63.3887 0.0300

Front-lit Image 60.7710 0.0549

5. Conclusions This study has presented a modified contrast compensation method by fuzzy logic classification, global thresholding and image illumination analysis for color face images. The input images can be either noise free or noisy face images. The input images are classified by fuzzy logic classification method into back-lit, normal-lit, and front-lit first. Then, the illumination of the input image was analyzed to obtain the image illumination distributions. Illumination distribution is found by using histogram processing. Each distribution is represented by one peak and two valleys. These peaks and valleys are used to be the parameters of the piecewise linear transformation. The implemented method is tested on color and gray level face images with and without noise. Performance analysis indicated that this method is efficient and effective by comparing with histogram equalization (HE) and mean preserving bi-histogram equalization (BBHE).

References [1] C. M. Tsai and Z. M. Yeh, “Contrast Enhancement by Automatic and Parameter-Free Piecewise Linear Transformation for Color Images,” IEEE Transactions on Consumer Electronics, vol. 54, no. 2, pp. 213- 219, 2008 [2] C. C. Sun, S. J. Ruan, M. C. Shie, and T. W. Pai, “Dynamic contrast enhancement based on histogram specification,” IEEE Transactions on Consumer Electronics, vol. 51, no. 4, pp. 1300-1305, 2005 [3] M. Abdullah-Al-Wadud, Md. H. Kabir, M. A. Akber dewan, O. Chae, “A Dynamic Histogram Equalization for Image Contrast Enhancement,” IEEE Trans. Consumer Electronics, vol. 53, no. 2, pp. 593 – 600, May 2007 [4] D. Menotti, L. Najman, J. Facon, and A. de A. Araujo, “Multihistogram equalization methods for contrast enhancement and brightness preserving,” IEEE Trans. Consumer Electronics, vol. 53, no.3, pp. 1186–1194, August 2007 [5] B. R. Lim, R. H. Park, and S. H. Kim, “High Dynamic Range for Contrast Enhancement,” IEEE Trans. Consumer Electronics, vol.52, no. 4, pp. 1454–1462, Nov. 2006 [6] L. Meylan and S. Süsstrunk, “Color image enhancement using a retinexbased adaptive filter,” In Proc. IS&T Second European Conference on Color in Graphics, Image, and Vision (CGIV 2004), 2, pp. 359363,2004 [7] S.-D. Chen and A. Ramli, “Minimum mean brightness error bihistogram equalization in contrast enhancement,” IEEE Trans. on Consumer Electronics, vol. 49, no. 4, pp. 1310-1319, Nov. 2003 [8] R. C. Gonzalez and R. E. Woods, Digital Image Processing, 3rd Ed., Prentice Hall, 2008 [9] Y.-T. Kim, “Contrast enhancement using brightness preserving bihistogram equalization,” IEEE Trans. on Consumer Electronics, vol. 43, no. 1, pp. 1-8, Feb. 1997 Noel Thomas,IJRIT

314

IJRIT International Journal of Research in Information Technology, Volume 2, Issue 6, June 2014, Pg: 308-315

[10] J.-S. R. Jang and C.-T. Sun, Neuro-Fuzzy and Soft Computing: A Computational Approach to Learning and Machine Intelligence, Prentice Hall, 1997 [11] Mariano Rivera, Oscar Dalmau, Washington Mio And Alonso Ramirez-Manzanares “Spatial Sampling for Image Segmentation” in The Computer journal,2007 [12] S. Lau, “Global image enhancement using local information,” Electronics Letters, vol. 30, pp. 122– 123, Jan. 1994 [13] Y. Q. Li, "Application of adaptive histogram equalization to X-ray chest image," Proc. of the SPIE 2321: 513-514, 1994 [14] S. M. Pizer et al., "Adaptive histogram equalization and its variations,'' Computer Vision, Graphics and Image Processing, 39:355-368, 1987 [15] J. Zimmerman, S. Pizer, E. Staab, E. Perry, W. McCartney, and B. Brenton, “Evaluation of the effectiveness of adaptive histogram equalization for contrast enhancement,” IEEE Tr. on Medical Imaging, pp. 304-312, Dec. 1988 [16] Yu Wan, Qian Chen and Bao-Min Zhang., “Image Enhancement Based On Equal Area Dualistic SubImage Histogram Equalization Method,” IEEE Trans Consumer Electronics, vol. 45, no. 1, pp. 68-75, Feb. 1999 [17] C. Munteanu and A. Rosa, “Color image enhancement using evolutionary principles and the retinex theory of color constancy,” In Neural Networks for Signal Processing XI, 2001. Proceedings of the 2001 IEEE Signal Processing Society Workshop, pp. 393-402, 2001

Noel Thomas,IJRIT

315