novel contrast enhancement scheme for infrared image using detail-preserving stretching

Upload: mohamed-taha

Post on 08-Mar-2016

227 views

Category:

Documents


1 download

DESCRIPTION

Research Paper Research paper in Computer Science

TRANSCRIPT

  • Novel contrast enhancement scheme forinfrared image using detail-preservingstretching

    Jin-Hyung KimJun-Hyung KimSeung-Won JungChang-Kyun NohSung-Jea Ko

    Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 05/11/2015 Terms of Use: http://spiedl.org/terms

  • Optical Engineering 50(7), 077002 (July 2011)

    Novel contrast enhancement scheme for infraredimage using detail-preserving stretchingJin-Hyung KimJun-Hyung KimSeung-Won JungKorea UniversitySchool of Electrical EngineeringAnamdong 5-gaSeongbuk-guSeoul, 136-713, Republic of Korea

    Chang-Kyun NohAgency for Defense DevelopmentThe 3rd R&D Institute-1Yuseong P. O. Box 35Daejeon, 305-600, Republic of Korea

    Sung-Jea KoKorea UniversitySchool of Electrical EngineeringAnamdong 5-gaSeongbuk-guSeoul, 136-713, Republic of KoreaE-mail: [email protected]

    Abstract. A novel contrast enhancement scheme for infrared (IR) imageusing detail-preserving stretching (DPS) is presented in this paper. Un-like CCD images, the histogram of the IR images is not distributed tothe entire dynamic range of the input image. Therefore, in order to en-hance the contrast, the proposed method adopts contrast stretching sinceit has a low computational complexity and preserves the global contrastof the original. However, the stretching often produces unpleasant imagesthat do not contain clear details by limiting the output dynamic range. Tosolve this problem, we propose a new contrast stretching method basedon gradient-domain processing (GDP) that can enhance the local imagecontrast while preserving the global contrast as well as the image details.Experimental results demonstrate that the proposed method outperformsthe conventional stretching-based and other image enhancement meth-ods with respect to the visual quality and computational complexity. C 2011Society of Photo-Optical Instrumentation Engineers (SPIE). [DOI: 10.1117/1.3597639]

    Subject terms: contrast enhancement; infrared; contrast stretching; gradient-domain processing.

    Paper 110122R received Feb. 9, 2011; revised manuscript received May 4, 2011;accepted for publication May 17, 2011; published online Jul. 6, 2011.

    1 IntroductionAn infrared (IR) image provides the visual information thatcannot be detected by the human eye. Therefore, it has beenwidely used for not only a military purpose, but also, a civilianpurpose.1 In the early days, the civilian application usingthe IR image had been limited to surveillance and weatherforecasting. Recently, it is expanded and adopted in consumerdevices such as the surveillance camera, night vision systemin a vehicle, mobile phone with infrared video, and personalhand-held camera.

    Unlike other images captured by the CCD/CMOSsensors, the IR image typically has low contrast because theIR image sensor cannot clearly differentiate the object fromthe background if they have a similar emissivity, especiallyfor long-wave IR images. Therefore, a contrast enhancementtechnique is essential to improve visibility of IR images.In the literature, several image enhancement techniqueshave been proposed to improve the contrast of the image.One of the most popular contrast enhancement techniquesis histogram equalization (HE).2 HE is widely used due toits low computational complexity and affordable contrastenhancement capability. However, since the histogramdistribution of the IR image is different from that of theCCD/CMOS image, it often over-enhances the image, whichmeans HE is not applicable to the IR image.3

    Various methods have been proposed to improve theperformance in terms of the visual quality, most of which areobtained by modifying HE, for example, bi-HE,4 dualisticsub-image HE,5 multi-HE,6 histogram modification frame-

    0091-3286/2011/$25.00 C 2011 SPIE

    work (HMF),7 contrast limited adaptive HE (CLAHE),8 etc.Bi-HE, dualistic sub-image HE, and multi-HE can preventthe change of the mean brightness of the input image but theycannot control the level of the enhancement. On the contrary,HMF can control the level of the enhancement, but itchanges the mean brightness of the input image.9 Comparedto other HE-based techniques, CLAHE can enhance thecontrast of the IR image, including a large low contrast areaproduced by the similar brightness level. However, it tendsto over-enhance the contrast and thus produce unnaturalimages.10

    Other efficient contrast enhancement methods are the hu-man visual system (HVS)-based methods such as perception-based contrast enhancement (PCE)11 and adaptive counter-shading (ACS).12 These methods utilize the HVS factors,such as the contrast sensitivity and luminance masking, toimprove the local image contrast. Although they enhance thelocal contrast clearly, the global contrast can be distortedby local processing. For IR images, in order to recognizethe target object correctly, it is required that the relative in-tensity level between the target and background objects isnot changed, which means the global contrast needs to bepreserved.

    To design the contrast enhancement method which canpreserve the global contrast and produce the natural image,we notice contrast stretching.2, 13 The contrast stretching isone of the simple image contrast enhancement techniquesthat can preserve the global contrast and produce a naturalimage. Moreover, it can be applied to the system requiringthe real-time processing due to the low computational com-plexity. However, the detailed information of the originalimage can be lost, and it degrades the visual quality of the

    Optical Engineering July 2011/Vol. 50(7)077002-1

    Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 05/11/2015 Terms of Use: http://spiedl.org/terms

  • Kim et al.: Novel contrast enhancement scheme for infrared image...

    Table 1 A brief description (includes strengths and weakness) for the conventional methods.

    Description Strengths Weakness

    HE-based Utilizes the local histograms Preserves the mean brightness Cannot control the level of theenhancement

    HMF Defines a functional to adjust amapping function

    Can control the level of theenhancement

    Changes the mean brightness

    CLAHE Limits the contrast level Enhances the contrast whichincludes a large flat area

    Tends to overenhance

    HVS-based Utilizes the HVS factors Enhances the local contrastclearly

    Distorts the global contrast

    Stretching Expands the dynamic range ofthe original image

    Fast, preserves the globalcontrast

    Tends to distort the details of theoriginal

    resultant image. Table 1 represents a brief description for theconventional methods.

    In this paper, we propose a novel contrast enhancementscheme which can preserve the global contrast and the de-tailed information of the original simultaneously by combin-ing the contrast stretching and a gradient-domain processing(GDP).14 The remainder of this paper is organized as fol-lows. Section 2 introduces the proposed contrast enhance-ment scheme. Experimental results and analysis are given inSec. 3 and the conclusion is drawn in Sec. 4.

    2 Proposed Contrast Enhancement SchemeAs explained in Sec. 1, contrast stretching is a very usefulmethod to enhance the contrast of IR imagery due to its char-acteristic of the histogram. However, it can cause the loss ofthe detail in high or low luminance area due to the limitationof the output range to be stretched. The proposed methodalso adopts the contrast stretching method, but the GDP isnewly addressed to cope with the detailed loss problem. Inthis section, we first review the conventional stretching andthen a detailed description of the proposed method nameddetail-preserving stretching (DPS) follows.2.1 Contrast StretchingContrast stretching is one of the simple image enhancementtechniques. It stretches the original range of pixel valuesto a desired range. Since the histogram of an IR image ismostly concentrated in certain parts (mostly near the areacorresponding to the ambient temperature10) as shown inFig. 1(d), the stretching can enhance the contrast of IRimage effectively as shown in Fig. 1(c) whereas HE resultsin the unnatural looking image as shown in Fig. 1(b). Thestretched pixel value p(x) for the original pixel value p(x)is determined as follows:

    p(x) =

    ph, if p(x) > ph,K [p(x) pl] + pl , if pl p(x) ph,pl, if p(x) < pl ,

    (1)

    where x is a pixel in an image, K is a stretching gain definedby

    K = ph plph pl , (2)

    where ph , pl , ph , and pl denote the upper, lower, stretchedupper and stretched lower pixel value limits, respectively. As

    described in Eq. (1), values below pl are set to pl and valuesabove ph are set to ph . Figure 2 shows a typical mapping func-tion of the contrast stretching. The mapping function is mono-tonically increasing, which means that the relative intensitylevel of the pixel values is not inverted by the stretching.Thus, the global contrast of the original image is preserved.

    The performance of the stretching is dependent on howpixel value limits are set. Among many stretching methods,the simplest one is to normalize the pixel value range fromthe minimum to the maximum of the dynamic range of theimage. To be specific, ph , pl , ph , and pl are set to the highestpixel value, lowest one, 2N 1, and 0 for N -bit gray-levelimage, respectively. This stretching, linear stretching, isvery simple and provides proper visual quality, however, asmall area with a very high or low gray-level [in a circle atthe top-right of Fig. 3(a)] can severely affect the value ofpl and ph , which could lead to unrepresentative scaling asshown in Fig. 3(b).

    A more robust approach to avoid this problem is mean-based stretching. The mean-based stretching selects the lim-its using the mean and standard deviation of the pixel valuesin the input image. Specifically, the limits pl and ph are setas follows:

    pl = m , ph = m + , (3)where m and are the mean and standard deviation of theoriginal image. By properly choosing the positive coefficients and , the mean-based stretching prevents outliers affectingthe scaling, and thus the global contrast can be preserved andthe resultant image is visually pleasing, as shown in Fig. 3(c).

    However, the mean-based stretching has a critical draw-back. Since the pixel values below pl and above ph are setto pl and ph respectively, it is observed that the mean-basedstretching causes the loss of the detail in the high or low lu-minance area as shown in Fig. 4(a) , and the artifact is namedsaturation.

    2.2 Detail-Preserving Stretching UsingGradient-Domain Processing

    To prevent the saturation and preserve the details of the orig-inal, the proposed method utilizes image gradients of theoriginal image. In the literature, the image gradients havebeen widely used for the local contrast enhancement11, 12, 15and the image editing.16 Especially, Lu et al. proposed thecontrast stretching using multiscale gradients. However, the

    Optical Engineering July 2011/Vol. 50(7)077002-2

    Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 05/11/2015 Terms of Use: http://spiedl.org/terms

  • Kim et al.: Novel contrast enhancement scheme for infrared image...

    Num

    berof

    pixe

    ls

    (c)

    (f)

    GraylevelGraylevel

    (a) (b)

    (d) (e)

    Num

    berof

    pixe

    ls

    Num

    berof

    pixe

    ls

    Graylevel0 0.2 0.4 0.6 0.8 1

    Fig. 1 Mountain: (a) Original, (b) histogram equalization, (c) the linear stretched, (d) the histogram of (a), (e) the histogram of (b), (f) the histogramof (c).

    method works only on the wavelet transform, and the detailloss due to the contrast stretching is not considered. To de-sign a detail-preserving contrast enhancement method dedi-cated to IR image processing, we adopt the GDP, the recentlydeveloped and most popular method, to manipulate the gra-dient and the pixel values of the image.14 Figure 5 showsthe block diagram of the proposed method. First, by apply-ing the mean-based stretching, the stretching gain K and thestretched image v are obtained, and then the gradient of theoriginal image g is computed. By using K and g, the targetgradient G is calculated, and then the gradient field of thestretched image is replaced to G. Last, the GDP finally recon-structs the resultant image f using G and v . The followingdescribes the detailed procedure.

    From Eq. (1), if both pixel values of the ith and (i + 1)thpixels exist within the limits, which means the pixel value

    Input gray level

    Out

    put g

    ray

    leve

    l

    0

    2N-1

    2N-1hplp

    'hp

    'lp

    Fig. 2 Mapping function of the contrast stretching.

    is not saturated by the stretching, the corresponding gradientG(xi ) of the ith pixel in the stretched image is obtained byG(xi ) = p(xi ) p(xi+1)

    = K [p(xi ) pl] K [p(xi+1) pl]= K [p(xi ) p(xi+1)] = K g(xi ), (4)

    where g(xi ) denotes the gradient of the ith pixel in theoriginal image and the gradient is defined as a differencebetween two adjacent pixels as follows:g (xi ) = p (xi ) p (xi + 1)

    = [p (xi , yi ) p (xi + 1, yi ) , p (xi , yi )p (xi , yi + 1)]. (5)

    Equation (4) means the gradient of the stretched imageis the same as that of the original image multiplied by K .Otherwise, if either p(xi ) or p(xi+1) exists beyond the limits,G(xi ) is less than g(xi ) multiplied by K or equal to zero. Wenotice that this is the loss of the detailed information by thestretching. To prevent this, we set G(xi ) to g(xi ) multipliedby K for all the pixels in the image. As a result, G(x) isdefined as follows:xi f

    {G (xi )} =xi f

    {K g (xi )}, (6)

    or equivalently

    G = K g. (7)By setting the above, the loss of the detailed information

    can be prevented. However, forcing the gradient of f toalways have G does not guarantee the preservation of theglobal image contrast. Therefore, we additionally constrain

    Optical Engineering July 2011/Vol. 50(7)077002-3

    Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 05/11/2015 Terms of Use: http://spiedl.org/terms

  • Kim et al.: Novel contrast enhancement scheme for infrared image...

    Nu m

    ber

    ofpi

    x el s

    Num

    ber

    ofpi

    xel s

    GraylevelGraylevel

    (a) (b)

    (d) (e)

    Num

    ber

    ofpi

    xels

    (c)

    (f)

    Graylevel0.2 0.4 0.6 0.8 10

    Fig. 3 The IR image including small and bright area: (a) Original; (b) the linear stretched; (c) the mean-based stretched ( = 3, = 1); (d) thehistogram of (a); (e) the histogram of (b); and (f) the histogram of (c).

    f to have pixel values similar to v because the global contrastof the original is maintained in v . To this end, we obtain fby minimizing the following energy function:

    x f

    E =x f

    (Ed + Eg), (8)

    where E is the energy function consisting of the data costfunction Ed and the gradient cost function Eg These func-tions are defined as follows:

    Ed = d ( f v)2 ,Eg = g{( fx Gx )2 + ( fy Gy)2}, (9)

    Graylevel

    (a) (b)

    (d) (e)

    (c)

    (f)

    Num

    berof

    pixe

    l s

    Num

    berof

    pixe

    ls

    Num

    berof

    pixe

    ls

    1 0 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 10

    Graylevel Graylevel

    Fig. 4 The IR image: (a) Original; (b) the linear stretched; (c) the mean-based stretched ( = 3, = 1); (d) the histogram of (a); (e) the histogramof (b); (f) the histogram of (c).

    Optical Engineering July 2011/Vol. 50(7)077002-4

    Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 05/11/2015 Terms of Use: http://spiedl.org/terms

  • Kim et al.: Novel contrast enhancement scheme for infrared image...

    Calculategradient field

    Gradient domainprocessing

    Stretching

    Stretched image

    Original image

    Enhanced image

    stretchinggain (K)

    X

    v

    gG

    Fig. 5 Flowchart of the proposed contrast enhancement algorithm.

    where fx and fy denote the horizontal and vertical derivativeof the final image f , Gx , and Gy respectively denote thehorizontal and vertical components of G, respectively. dand g are the weights for the data and the gradient costfunctions, respectively. The weights d and g control theamount of influence on the final image, and a method isintroduced on how to set the value of the weights.17 In thispaper, the weights are empirically set to 1 and this results inthe affordable visual quality.

    The final image f that minimizes Eq. (8) satisfies theEulerLagrange equation,

    E f

    x

    E fx

    yE fy = 0. (10)

    Rearranging Eq. (10) yields(d/g) f ( fxx + fyy) = (d/g)v (Gx + Gy), (11)

    Image 1 Image 2 Image 3 Image 4

    Image 5 Image 6 Image 7 Image 8

    Image 9 Image 10 Image 11 Image 12

    Image 13 Image 14 Image 15 Image 16

    Image 17 Image 18 Image 19 Image 20

    Fig. 6 The sample IR images.

    Optical Engineering July 2011/Vol. 50(7)077002-5

    Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 05/11/2015 Terms of Use: http://spiedl.org/terms

  • Kim et al.: Novel contrast enhancement scheme for infrared image...

    (a) (b)

    (d) (e)

    (c)

    (f)

    (g) (h) (i)

    Fig. 7 The mean-based stretched image with the limits pl and ph set to m and m + , respectively. and are (a) 1, 1, (b) 1, 2, (c) 1, 3,(d) 2, 1, (e) 2, 2, (f) 2, 3, (g) 3, 1, (h) 3, 2, and (i) 3, 3.

    or equivalently

    (d/g) f 2 f = (d/g)v G, (12)where fxx and fyy are the horizontal and vertical compo-nents of 2 f , the Laplacian of f , respectively, and Gdenotes the divergence of G. Equation (12) is a screenedPoisson equation, and finding f is equivalent to solving thescreened Poisson equation. To solve the equation, many so-lutions have been proposed, such as the conjugate-gradientmethod,18 a fast Fourier-domain solver,17 etc. These solutionsare fast enough for real-time processing and the conjugate-gradient method is adopted in the proposed method. Finally,by applying the conjugate-gradient method, the final resultf is obtained which minimizes E . For a detailed descrip-tion of the conjugate-gradient method, the reader may re-fer to Numerical Recipes in C (Ref. 19) or Shewchukspaper.18

    3 Experimental Results and AnalysisIn this section, we verify the performance of the proposedcontrast enhancement scheme. In the simulation, we havetested 20 IR images, where the resultant images obtainedby the proposed algorithm are provided in Fig. 6. Amongthem, the experiments on I mage8, I mage10, and I mage12with a resolution of 320240 including complex and tex-tured areas (building, trees, etc) are detailed in this paper.Figure 7 illustrates the mean-based stretched image of thesample image 12 with varying the coefficients and . Ascan be seen, the visual quality of the mean-based stretched

    image is dependent on the coefficients. We set and to 3and 1 experimentally, which produce a proper visual qualityfor the sample IR images.

    To verify the performance of the detail-preserving,Fig. 8 shows the comparison of the gradient fields of thesample image 8 between the mean-based stretched imageand the resultant image obtained by the proposed method. It

    (a)

    1

    2

    3

    1

    2

    3

    Mean-based stretched Proposed1 1

    2 2

    3 3

    (b)

    Fig. 8 (a) The horizontal gradient fields of the mean-based stretchedimage (top) and the resultant image with the proposed method (bot-tom). (b) The enlarged sub-images of each numbered area.

    Optical Engineering July 2011/Vol. 50(7)077002-6

    Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 05/11/2015 Terms of Use: http://spiedl.org/terms

  • Kim et al.: Novel contrast enhancement scheme for infrared image...

    Fig. 9 Histograms of the original image (top) and the resultant images with the mean-based stretching (middle) and the proposed scheme(bottom). (a) Image8, (b) Image10, and (c) Image12.

    (a) (b)

    (d) (e)

    (c)

    (f)

    Fig. 10 Image 8: (a) Original; (b) the mean-based stretching, (c) perception-based contrast enhancement; (d) CLAHE; (e) adaptive counter-shading; (f) the proposed.

    Optical Engineering July 2011/Vol. 50(7)077002-7

    Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 05/11/2015 Terms of Use: http://spiedl.org/terms

  • Kim et al.: Novel contrast enhancement scheme for infrared image...

    (a) (b)

    (d) (e)

    (c)

    (f)

    Fig. 11 Image 10: (a) Original; (b) the mean-based stretching; (c) perception based contrast enhancement, (d) CLAHE; (e) adaptive counter-shading; (f) the proposed.

    can be observed that the detailed information is more clearlypreserved in the proposed method as compared with thatin the mean-based stretched image. By comparing the his-tograms in Fig. 9, it is seen that the data loss caused bysaturation is reduced. To be specific, the proposed methodreduces the number of the outlier pixels denoted by the cir-

    cles in Fig. 9 by 46.4%, 44.6%, and 60% for the sampleimage 8, 10, and 12 respectively, compared to the mean-based stretching.

    In Figs. 10, 11, and 12, we compare the resultant imagesobtained by the mean-based stretching, CLAHE, perception-based contrast enhancement, adaptive counter-shading, and

    (a) (b)

    (d) (e)

    (c)

    (f)

    Fig. 12 Image 12: (a) Original; (b) the mean-based stretching; (c) perception-based contrast enhancement; (d) CLAHE; (e) adaptive counter-shading; (f) the proposed.

    Optical Engineering July 2011/Vol. 50(7)077002-8

    Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 05/11/2015 Terms of Use: http://spiedl.org/terms

  • Kim et al.: Novel contrast enhancement scheme for infrared image...

    Table 2 Processing speed for the sample images (milliseconds).

    Image MS PCE ACS CLAHE Proposed

    1 12.4 584.2 80.3 341.9 22.4

    2 12.7 1091.8 80.7 335.9 22.7

    3 12.5 1084.7 80.7 346.2 22.4

    4 12.6 520.1 81.1 344.6 26.8

    5 12.7 1099.3 81.0 354.3 23.5

    6 12.8 550.4 81.7 342.1 22.8

    7 13.3 679.0 85.2 337.9 23.4

    8 12.6 2364.7 81.4 342.1 23.3

    9 12.7 444.1 81.1 340.3 22.6

    10 12.5 1598.1 81.3 340.5 22.8

    11 12.7 548.8 81.6 335.8 22.8

    12 12.8 1214.7 85.1 342.0 22.7

    13 12.6 481.2 80.3 343.4 26.3

    14 12.5 904.5 81.2 342.2 22.6

    15 12.8 2034.6 81.2 340.8 22.7

    16 12.6 2128.4 85.8 342.6 22.3

    17 12.7 2294.4 85.1 338.4 22.1

    18 12.5 1058.7 81.5 341.9 22.7

    19 12.6 362.2 81.0 336.9 22.7

    20 12.6 553.6 86.9 342.0 24.8

    Average 12.7 1079.9 82.2 341.6 23.2

    Note: MS is mean-based stretching.

    the proposed method. It is observed that the mean-basedstretching preserves the global contrast, but the details arelost. The perception-based contrast enhancement fails tocontrol the brightness and the adaptive countershading dis-torts the global contrast. The resultant images obtained byCLAHE show un-natural looking images. The resultant im-ages obtained by the proposed scheme demonstrate that theproposed method effectively improves the visual quality, aswell as preserves the global contrast and the detail.

    In addition, we evaluate the computational complexityof the proposed method. Comparison results of the com-putational complexity for the conventional and proposedmethods are summarized in Table 2. As compared with themean-based stretching, the proposed method spends moretime in the computation since the scheme performs the GDPadditionally. Except this case, it is noted that the proposedmethod achieves the higher performance in terms of theprocessing speed.

    Last, we compare the performance of the global contrastpreserving using block mean error (BME) between the con-ventional methods and the proposed one. The BME R is

    Table 3 Block mean error of the sample images.

    Image MS PCE ACS CLAHE Proposed

    1 0.19 0.01 0.05 0.11 0.05

    2 0.01 0.11 0.04 0.20 0.01

    3 0.05 0.03 0.04 0.11 0.05

    4 0.12 0.09 0.07 0.20 0.02

    5 0 0.09 0.15 0.12 0

    6 0.11 0.09 0.07 0.14 0.01

    7 0.09 0.05 0.05 0.18 0

    8 0.02 0.11 0.08 0.06 0.02

    9 0.09 0.05 0.06 0.10 0

    10 0.02 0.15 0.07 0.09 0

    11 0.14 0.14 0.05 0.21 0.01

    12 0.07 0.13 0.10 0.08 0.01

    13 0.07 0.02 0.13 0.11 0.01

    14 0.05 0.16 0.08 0.15 0.02

    15 0.07 0.05 0.07 0.15 0.01

    16 0.03 0.09 0.06 0.09 0

    17 0.03 0.08 0.07 0.08 0

    18 0.11 0.02 0.07 0.16 0.01

    19 0.03 0.07 0.07 0.21 0.01

    20 0.03 0.08 0.06 0.21 0.01

    Average 0.07 0.08 0.07 0.14 0.01

    obtained by the followings:

    R( f ) =bk f

    |Mo(bk) Mt (bk)|, (13)

    M(bk) = [m(bk) mmin]/(mmax mmin), (14)where m(bk) is the mean of the pixel value of the kth blockbk in the image. mmin and mmax denote the minimum andmaximum value of the mean. M(bk) is a normalized meanvalue, Mo and Mt denote the means of the original and target,respectively. If R is close to zero, the global contrast of thetarget image is similar to the one of the original, which meansthe global contrast of the original is preserved. In Table 3, itis observed that the BME of the proposed method is smallerthan others.

    4 ConclusionIn this paper, we proposed the new contrast enhancementscheme for the IR image using detail-preserving stretching.The proposed method is posed as a minimization problemthat balances preservation of the gradient with preservation

    Optical Engineering July 2011/Vol. 50(7)077002-9

    Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 05/11/2015 Terms of Use: http://spiedl.org/terms

  • Kim et al.: Novel contrast enhancement scheme for infrared image...

    of the global contrast. It utilizes the contrast stretching toimprove a visual quality of the input image; however, thestretching can cause the loss of the detail part in an image.To cope with this problem, DPS was newly presented to pre-serve the detail and enhance the contrast of the image usingGDP. Experimental results show that the proposed methodcan produce a natural looking image by enhancing the lo-cal contrast while preserving the global contrast, as well asthe image details. Since the proposed DPS method improvesthe computing speed as compared with the others, it can besuccessfully adopted in the hand-held camera, the surveil-lance system, and driving assistant system in a vehicle whichrequires real-time processing.

    AcknowledgmentsThis research was supported by the Agency for Defense De-velopment (ADD) under the project Local area processingfor EO/IR image and by Mid-career Researcher Programthrough a National Research Foundation of Korea (NRF)grant funded by the Korea government (MEST) (No. 2011-0000200).References

    1. A. Bovik, Handbook of Image and Video Processing, 2nd ed., ElsevierAcademic Press, New York (2005).

    2. R. Gonzalez and R. Woods, Digital Image Processing, 2nd ed., PrenticeHall, Englewood cliffs (2002).

    3. R. Lai, Y.-T. Yang, B.-J. Wang, and H.-X. Zhou, A quantitative measurebased infrared image enhancement algorithm using plateau histogram,Opt. Commun. 283, 42834288 (2010).

    4. Y. T. Kim, Enhancement using brightness preserving bi-histogramequalization, IEEE Trans. Consum. Electron. 43(1), 18 (1997).

    5. Y. Wang, Q. Chen, and B. Zhang, Image enhancement based on equalarea dualistic sub-image histogram equalization method, IEEE Trans.Consum. Electron. 45(1), 6875 (1999).

    6. D. Menotti, L. Najman, J. Facon, and A. de A. Araujo, Multi-histogramequalization methods for contrast enhancement and brightness preserv-ing, IEEE Trans. Consum. Electron. 53(3), 11861194 (2007).

    7. T. Arici, S. Dikbas, and Y. Altunbasak, A histogram modificationframework and its application for image contrast enhancement, IEEETrans. Image Process. 18(9) 19211935 (2009).

    8. A. M. Reza, Realization of the contrast limited adaptive histogramequalization (CLAHE) for real-time image enhancement, J. VLSI Sig-nal Proc. Syst. Signal Image, Video Technol. 38, 3544 (2004).

    9. C. Wang, J. Peng, and Z. Ye, Flattest histogram specification withaccurate brightness preservation, IET Image Process. 2, 249262(2008).

    10. F. Branchitta, M. Diani, G. Corsini, and A. Porta, Dynamic-rangecompression and contrast enhancement in infrared imaging systems,Opt. Eng. 47(7), 076401 (2000).

    11. A. Majumder and S. Irani, Perception-based contrast enhancement ofimages, ACM Trans. Applied Perception 4(3), (2007).

    12. G. Krawczyk, K. Myszkowski, and H.-P. Seidel, Contrast restora-tion by adaptive countershading, Proc. Eurographics 2007 (ComputerGraphics Forum), Vol. 26(3), pp. 581590, Blackwell, Prague, CzechRepublic (2007).

    13. J. Lu and D. M. Healy Jr., Contrast enhancement via multiscale gra-dient transformation, Proc. IEEE International Conference on ImageProcessing (ICIP)1994, Vol. 2, 482486, IEEE, Austin, TX (1994).

    14. P. Bhat, C. L. Zitnick, M. Cohen, and B. Curless, GradientShop: Agradient-domain optimization framework for image and video filtering,ACM Trans. Graphics 29(2), 10, 114 (2010).

    15. J. Lu, D. M. Healy Jr., and J. B. Weaver, Contrast enhancement ofmedical images using multiscale edge representation, Opt. Eng. 2(7),21512161, ACM, San Diego, CA (1994).

    16. P. Perez, M. Gangnet, and A. Blake, Poisson image editing, Proc.ACM SIGGRAPH 2003, Vol. 22, pp. 313318 (2003).

    17. P. Bhat, B. Curless, M. Cohen, and L. Zitnick, Fourier analysis of the2D screened poisson equation for gradient domain problems, Proc. theEuropean Conf. Computer Vision, Lecture Notes in Computer Science,Vol. 5303, pp. 114128, Springer, Marseille, France (2008).

    18. J. Shewchuk, An introduction to the conjugate gradient method with-out the agonizing pain, Technical Report, CMU-Cs-94-125, CarnegieMellon University, Pittsburgh, PA (1994).

    19. W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery,Numerical Recipes in C: The Art of Scientific Computing, CambridgeUniversity Press, New York, NY, chap. 2.7.6, pp. 8793 (1992).

    Jin-Hyung Kim received a BS degree inelectronic engineering from Korea Univer-sity in 2004. He is currently working to-ward his PhD degree in the Department ofElectronic Engineering at Korea University.His research interests are image enhance-ment, image compression, and digital videoprocessing. His current research topic is adevelopment of a high performance cam-era system and image enhancement algo-rithm for high resolution and high dynamicrange video.

    Jun-Hyung Kim received a BS degree fromKorea University, in Electronic Engineering,in 2006. He is currently pursuing a PhD de-gree in School of Electrical Engineering atKorea University. His research interests arein the areas of image processing.

    Seung-Won Jung received BS and PhD de-grees in electrical engineering in 2005 and2011, respectively, from Korea University,Seoul, Korea. He is now a post-doctoral re-search fellow in the Research Institute of In-formation and Communication Technology,Korea University, Seoul, Korea. His researchinterests include image enhancement, imagerestoration, video compression, and com-puter vision.

    Chang-Kyun Noh received an MS degree inelectronics engineering from Yonsei Univer-sity, Seoul, Korea, in 2000. He is currentlyworking at the Agency for Defense Develop-ment for developing EO/IR technology. Hiscurrent research topic is a development ofan EO/IR image enhancement algorithm.

    Sung-Jea Ko received a PhD degree in 1988and an MS degree in 1986, both in electricaland computer engineering, from State Uni-versity of New York at Buffalo, and a BS de-gree in electronic engineering at Korea Uni-versity in 1980. In 1992, he joined the De-partment of Electronic Engineering at KoreaUniversity where he is currently a professor.From 1988 to 1992, he was an assistant pro-fessor of the Department of Electrical andComputer Engineering at the University of

    MichiganDearborn. He has published over 140 international journalarticles. He also holds over 50 patents registered on video signalprocessing and multimedia communications. He is the 1999 Recipi-ent of the LG Research Award given to the Outstanding InformationTechnology and Communication Researcher. He received the Hae-Dong best paper award from the IEEK (1997) and the best paperaward from the IEEE Asia Pacific Conference on Circuits and Sys-tems (1996), and the research excellence award from Korea Univer-sity (2004). He has served as a TPC member for the IEEE Conferenceon Consumer Electronics (ICCE) since 1997 and received a 10-yearservice award from the TPC of ICCE in 2008. He is also a TPC mem-ber and an international advisor of ICCE-Berlin and a member of theeditorial board of the IEEE Transactions on Consumer Electronics.

    Optical Engineering July 2011/Vol. 50(7)077002-10

    Downloaded From: http://opticalengineering.spiedigitallibrary.org/ on 05/11/2015 Terms of Use: http://spiedl.org/terms