Colour Image Enhancement by Virtual Histogram Approach

Transcription

1 704 IEEE Transactions on Consumer Electronics, Vol. 56, No., May 010 Colour Image Enhancement by Virtual Histogram Aroach Zhengya Xu, Hong Ren Wu, Xinghuo Yu, Fellow, IEEE, Bin Qiu, Senior Member, IEEE Abstract This aer introduces a new hybrid image enhancement aroach driven by both global and local rocesses on luminance and chrominance comonents of the image. This aroach, based on the arameter-controlled virtual histogram distribution method, can enhance simultaneously the overall contrast and the sharness of an image. The aroach also increases the visibility of secified ortions or asects of the image whilst better maintaining image colour. The aroach was comared with other well-known image enhancement techniques. The exerimental results have shown the sueriority of the roosed aroach 1. Index Terms Image rocessing, image enhancement. I. INTRODUCTION Image enhancement, which transforms digital images to enhance the visual information within, is a rimary oeration for almost all vision and image rocessing tasks in several areas such as comuter vision, biomedical image analysis, forensic video/image analysis, remote sensing and fault detection [, 4]. For examle, in forensic video/image analysis tasks, surveillance videos have quite different qualities comared with other videos such as the videos for high quality entertainment or TV broadcasting. High quality entertainment or broadcasting videos are roduced under controlled lighting environment, whereas surveillance videos for monitoring outdoor scenes are acquired under greatly varied lighting conditions deending on the weather and the time of the day. One of the common defects of surveillance videos is oor contrast resulting from reduced image brightness range. A routine examination of the histograms of the images from the videos reveals that some of the images contain relatively few levels of brightness, and some of the images have a tye of histograms. In the tye of histograms, a large san of the intensity range at one end is unused while the other end of the intensity scale is crowded with high frequency eaks [4], which is tyically reresentative of imroerly exosed images. The roblem is how to aroximate or reconstruct information that was lost because of the image having been catured under subotimal aerture or exosure conditions. Enhancement transformation to modify the contrast of an image within a dislay s dynamic range is, therefore, required in order to 1 Z. Xu, H.R. Wu and X. Yu are with Royal Melbourne Institute of Technology, University, VIC 3001, Australia ( zhengya.xu@rmit.edu.au; henry.wu@rmit.edu.au; x.yu@rmit.edu.au). B. Qiu is with Faculty of Information Technology, Monash University, VIC 3800, Australia ( bin.qiu@infotech.monash.edu.au). Contributed Paer Manuscrit received November 18, 009 Current version ublished ; Electronic version ublished reveal full information contents in the videos, e.g., for forensic investigations. First of all, the basic strategies are briefly reviewed for image enhancement. Point-oeration-based image enhancement includes contrast stretching, non-linear oint transformation and histogram modelling [3, 4]. They are zero memory oerations that rema a given inut grey-level into an outut grey-level, according to a global transformation [, 4, 9]. Non-linear oint transformations, which could be exressed as G(j,k)=[F(j,k)] where F(j,k) reresents the original image, G(j,k) reresents the outut image and is the ower law variable, have been shown to imrove visual contrast in some cases whilst clearly imairing visual contrast in other cases [, 3, 15]. In histogram modelling [10, 13, 17,, 4], the original image is scaled so that the histogram of the enhanced image is forced to be some desired form such as uniform, exonential, hyerbolic or logarithmic [18, 19]. These methods have the disadvantage of treating the image globally only. In order to differentiate between several areas of the image that may require different levels of contrast enhancement, an adative histogram modelling technique was roosed [16]. Images generated by the adative histogram modelling rocess, sometimes, is so harsh on the image visual aearance that an adative blurring of the window histogram was roosed [14] rior to forming the cumulative histogram as a means of imroving the image quality. Recently, some histogram based aroaches, such as dynamic range searate histogram equalization (DRSHE)[6], brightness reserving dynamic histogram equalization (BPDHE)[7] and gain-controllable clied histogram equalization (GC-CHE)[8] have been develoed in order to overcome some drawbacks of histogram equalization methods. Other classes of methods for image enhancement are aroaches based on the Retinex theory [1, 3], satial oerations and seudo-colouring. Satial oerations may suffer from enhancing excessively the noise in the image or conversely smoothing the image in areas that need to reserve shar details [3] and these oerations are also known to be time consuming. Pseudo-colouring methods artificially ma the grey-scale image to a colour image, with the disadvantage that extensive interactive trials are required to determine an accetable maing scheme []. Local enhancement methods have been develoed based on the gray-level distribution in the neighbourhood of every ixel in a given image. A tyical examle of local enhancement methods is the adative histogram equalization (AHE), which has shown good results in medical imaging alications. However, AHE uses an enhancement kernel /10/$ IEEE

2 Z. Xu et al.: Colour Image Enhancement by Virtual Histogram Aroach 705 that is quite comutationally exensive. Moreover, AHE may yield unsatisfactory oututs, e.g., images with noise artefacts and falsely enhanced shadows [1, 5]. Furthermore, all the aforementioned methodologies, excet the seud-colouring, only deal with grey-scale image enhancement, i.e., they only use luminance comonent of a colour image for colour image enhancement. With a esecial interest in surveillance video/image rocessing, the roosed colour image enhancement method is a fast adjustable hybrid aroach controlled by a set of arameters in order to take the advantages of oint oerations and local information driven enhancement techniques, in making effective use of the entire range of available ixel-values for both colour and luminance comonents of a colour image. The organization of this aer is as follows. Section II addresses the rinciles of the roosed image enhancement technique. Exerimental results are resented in Section III. Final conclusions are drawn in Section IV. II. PRINCIPLE OF THE PROPOSED METHOD In surveillance videos/images, the luminance histogram of a tyical natural scene that has been linearly quantised is, more often than not, highly skewed toward the darker levers; a majority of the ixels ossess a luminance less than the average. In such images, details in the darker regions are often not ercetible. One means to enhance these tyes of images is a technique called histogram modification, where the original image is scaled so that the histogram of the enhanced image follows a desired distribution. Usually, a uniform distribution is used to create an image with equally distributed brightness levels over the entire brightness scale. While histogram equalization alies a transformation that yields a close-to-uniform histogram for the relative frequency of the brightness-levels in an image, it only enhances the contrast for brightness values close to histogram maxima, and decreases the contrast near histogram minima [, 1, 15]. If the image analyst is interested in certain arts or features of an image and the brightness of the arts or features of the image is not close the histogram maxima or, even worse, near the histogram minima, which haens often in surveillance video/image analysis, histogram equalization is helless in the required contrast enhancement task. A linear-like or no-linear brightness stretching is only effective for an image where the histogram is narrow [, 9], which is, unfortunately, not often the case in ractical video/image analysis. In order to meet the above articular ractical demands and stringent requirements for forensic image/video analysis, biomedical image analysis and remote sensing, a new colour image enhancement method is roosed as follows. The roosed enhancement technique is driven by both global and local rocesses to achieve not only effective imrovement of overall contrast but also the significant enhancement of details in identified features/areas of interest of a colour image. The roosed method also aims at emloying a much less time-consuming enhancement mechanism than those used by the existing methods. Histograms are used to deict image statistics in an image interreted visual format. With histogram, it is easy to determine certain tyes of roblems in an image, such as if the image is roerly exosed. Luminance histogram and comonent histogram both rovide useful information about the lighting, contrast, dynamic range, and saturation effects relative to the individual colour comonents [4]. Therefore, in the roosed method the histogram of the enhanced colour image should not be saturated at one or both ends of the dynamic range or at least not bring new significant sikes at the tail ends in order not to introduce new oor exosure like defects in the image. In addition to a full use of the maximum ossible dynamic range, colour comonent and local information can, certainly, make a contribution to the contrast enhancement. Ideally, an effective image enhancement technique devised using colour comonent and local information must not have or introduce very large sikes in the histogram of the enhanced image. Therefore, the intention of the roosed method is to find a monotonic ixel brightness transformation q=t() for a colour image such that the desired outut histogram can not only meet secific requirements but also be as uniform as ossible over the whole outut brightness scale [9] to fill in the full range of brightness values. First of all, definitions and notations are resented for introduction of the roosed enhancement method. Let C { c ( c1 c ) 1 c1 M,1 c N} denote the, ixel coordinates of a colour image, where M and N are the height and width of the image, resectively. At each ixel coordinate, c C, a multivariate value x () [ (), (), ()] T RGB c xr c xg c xb c is used to reresent the ixel in RGB (Red, Green, Blue) colour sace at the current osition and a multivariate value T xyc () [ (), (), ()] BC c x R Y c xc c x B C c is used to reresent the R ixel in YC B C R colour sace [, 1]. For each RGB colour channel, each individual histogram entry is defined, resectively, as h R (i)=card{c xr () c =i, c C }, (1) h G (i)=card{c xg () c =i, c C }, () h B (i)=card{c xb () c =i, c C }, (3) where card{ } is the cardinality function, 0 i<k, and K is a scale for a comonent of the colour image and, usually, 56. Second, the colour image is, some times, transformed from the RGB colour sace or another colour sace to the YC B C R colour sace necessarily in the roosed image enhancement [, 1]. The luminance channel histogram of an image in the YC B C R colour sace is defined as h Y (i)=card{c xy ( c ) =i, c C }, (4)

3 706 IEEE Transactions on Consumer Electronics, Vol. 56, No., May 010 where all symbols are as defined in equations (1) to (3). The cumulative histogram for each RGB comonent and luminance comonent, Y, for the YC B C R colour sace are defined by extending the definition of cumulative histogram from grey-scale image, resectively as and R H ( ) h ( i), (5) H G B 0 0 R ( ) h ( i), (6) 0 G H ( ) h ( i) (7) Y B H ( ) h ( i). (8) Y 0 where the inut brightness value is [ 0, k ] and [ 0, k ]. The cumulative histograms are monotonic no-decreasing functions with H R (K)= H G (K)= H B (K)= H Y (K)=MN. (9) Based on above definitions, the roosed enhancement method is described as follows. Prominent image events, such as objects or a scene such as edges and contours, originated from local changes in intensity or colour, are highly imortant for visual ercetion and interretation of images. It is thus naturally that the enhancement of edges has been an imortant task in image rocessing. Comared with the original image, an enhanced image with good contrast will have a higher intensity of the edges. Since a histogram of an image contains no information about the satial arrangement of ixels in the image, luminance histogram and comonent histogram do not rovide any information about the satial distribution of the actual colours in the image. Since we are only interested in how to enhance the edge intensity without regard to its orientation, a linear differential oerator, which is a local geometric information based oerator, widely known as the Lalacian, x( c1, c ) x( c1, c ) x ( c1, c ) may be used in c c 1 order to enhance edge related area in colour images. In colour images, the scale of brightness is 56 reresented by 8-bit for most of images, whereas a true RGB colour sace has distinct 4 colours, with each colour comonent ixel reresented by 8-bit. Therefore, the edge information is not only described by their luminance but also conveyed by their colour. In order to use information fully from both of brightness and colour ercetion, the Lalacian oeration is alied to each of the RGB channels, resectively. Let L RGB (c) = ( c) + ( c) + ( c). (10) x R x G x B Based on (10), we can define S La ={ c L RGB (c) >T la, c C }, (11) where threshold T la [1, 10], and the default threshold T la is set to 3. Hence, S La is a set of ixel coordinates of an image, i.e., S La C. Each of the ixels with their coordinates in the set S La has a sum of absolute value of outut of the ixel rocessed with a Lalace oerator (10) and the sum value is greater than T la. We define h Yv ()=card{c xy ( c ) =, c S La }, (1) where the inut brightness value are [ 0, k ] and [ 0, k ]. h Yv ( i ) can be treated as a secial density function deended on the local feature of every ixel and the frequency of the brightness value of the ixels. If a histogram for an inut colour image is h Y () and the inut brightness value is [ 0, k ], H, H 1 and H, are defined as follows: H k 0 k1 k10 h i Y ( ) (13) H w h Y ( i) (14) 1 and v k H 0 w h i Yv( ) (15) where k1 and k10 are in the range of [ 0, k ]; h Yw ()= h() if is in the range of ( k10, k1 ], otherwise h Yw ()=0; w is a arameter with the default value of ; v is a arameter with the default value set to 1. Here, H 1 () is designed to suit secial enhancement requirements for the image interretation. H Using c n H H 1 H as a normalisation coefficient, a new virtual distribution function is defined as 0 h0 ( i) cn ( hy ( i) w hyw( i) vh 0 k10 Yv 0 ( i)) (16) If M and N are the height and the width of an image, resectively, and the outut brightness range is [q 0, q k ], the desired outut histogram can be aroximated with (16) by its corresonding continuous robability density as follows: MN 1 ds h s ds q 0() q0 qk q (17) 0 0 The left side of Equation (17) is the corresonding uniform robability distribution function. The desired ixel brightness histogram transformation T is defined as qk q q T( ) h ( s) dsq MN (18) 0

4 Z. Xu et al.: Colour Image Enhancement by Virtual Histogram Aroach 707 The discrete aroximation of the continuous ixel brightness transformation from Equation (18) is, therefore, given by qk q q T( ) h ( i) q MN (19) i 0 Thus, the quantisation ste-size is obtained as follows qk q qi MN vh ( )) Yv i 0 qk q0 h0 ( i ) MN c ( h( ) wh n i Yw ( i ) (0) On the right-hand side of Equation (0), the second term is used to enhance contrast for a secified range [P k10, P k1 ]; wh Yw (i) = 0, if i is not in the range of [P k10, P k1 ]; the third term, basically as the first term (inut histogram of the image), is deendent on the image structure, though the arameter v can be adjusted. In most cases, v is fixed as 1, since the enhanced result is not very sensitive to the change of the v (see Fig.1). Fig1.c and Fig1.d show the results of the tested image enhanced by the roosed method with different values of v. Through (0), it can be clearly seen how the outut interval value between adjacent two brightness values is roduced one by one and how the arameters make contribution to every outut brightness level for contrast enhancement since human vision is very sensitive to the interval value q [1, 0]. The default values of these arameters are: P k10 = 0, P k1 = 30 and w =. The arameters can be adjusted by an image interreter to meet his or her secific requirements. For many cases, the roosed aroach with the default values of the arameters works well without user intervention, as changes of the arameters do not affect the enhanced result very much (see Fig.1). Fig1.b Fig1.c and Fig1.d show the results of the tested image enhanced by the roosed method with different values of its arameters. It is noted that the number of reconstruction levels of the enhanced image must be less than or equal to the number of levels of original image to rovide roer intensity scale redistribution if all ixels in each quantisation level are to be treated similarly [1]. When the contrast of a dark area of an image whose histogram sans a broad range of the dislay scale is enhanced, the bright areas of the image may be out of the dislay range as a result of the above rescaling as defined by (19). When the contrast of a bright area of an image whose histogram sans a broad range of the dislay scale is enhanced, the dark areas of the image may also be out of the dislay range as a result of the above rescaling as defined by (19). Therefore, a hard-limit is needed to ma the outut image ixel values back into the dislay range [9]. However, the simle hard-limiting method is only suitable for an outut image with only a few ixels whose brightness values are outside [q 0, q k ]. In order to avoid or to greatly reduce the brightness range of the outut image, a rescaling constraint is emloyed using arameter t, which is introduced in the roosed method, to limit the maxima of q within t, and to smooth the enhancement contrast over the full brightness scale. Consider a atch of light of intensity I +I surrounded by a background of intensity I and I is actually similar to t for human vision. Over a wide range of intensities, it is found that the ratio thei/i, called the Weber fraction, is nearly constant at a value of about 0.0[], so the default value of t is set to 3. The rescaling rocess also relieves further the undesired roerty of the traditional histogram equalisation technique, which tends to reduce contrast near histogram minima. If an image with its histogram basically concentrated in a very bright region, the image can be first inversed. Then, the roosed method is alied and the resultant image is inversed back, to ensure above constraints being met and to work more effectively. After the brightness contrast enhancement in the luminance channel, the outut colour image is transformed back to the RGB colour sace for dislay, since almost all hardware generally deliver or dislay colour via the RGB channels. Though the limitation of q was alied, the outut results, unfortunately in some cases, showed that a histogram/histograms of RGB channels was/were saturated at one or both ends of the dynamic range. The out-of-range outut ixel values were maed into the maximum or minimum values of the outut scaling range and aeared on the histogram as significant sikes at the tail ends. They looked like what tyically occurs in an underexosed or overexosed image, which is undesirable for imroving visual quality of an image. In order to remove this defect, a more effective outut range boundary control mechanism is further alied in the roosed method. The outut range boundary control mechanism is introduced as follows. The linear maing of video signal from the RGB colour sace to the YC B C R colour sace [, 1], which is used by video and broadcasting television industry, is comuted for the luminance comonent, Y, by [, 1] Y=0.99R+0.587G+0.144B (1) where the luminosity (Y) is a function of R, G and B which are normalised to 1, and denoted as Y ( R, G, B). Y Y Obviously, according to (3), 0, 0, and R G Y 0. Hence, Y ( R, G, B) is a non-decreasing monotonic B function. In order to determine the new borders of the outut scaling range for the luminance channel (Y) we only need to find the corresonding Y values to the uer and the lower bounds of the RGB channels for the colour image. The values can be obtained from the outut RGB histograms and the conversion from the RGB sace to the YC B C R sace[, 15] is described as follows, Y low = max{ Y low-red, Y low-green, Y low-blue } ()

5 708 IEEE Transactions on Consumer Electronics, Vol. 56, No., May 010 and Y high = min{y high-red, Y high-green, Y high-red }, (3) where Y high-red = min{y y xy( c) ( xr( c ) 55) ( hr(55) hred old(55) s) } (4) Y high-green =min{y y x ( c) ( x ( c ) 55) ( h (55) h (55) s) } (5) Y G G greenold Y high-blue =min{y y xy ( c) ( xb ( c ) 55) ( hb (55) hblueold (55) s) } (6) Y low-red = max{y y xy () c ( xr () c 0) ( hr (0) hred old (0) s) }, (7) Y low-green = max{y y xy () c ( xg () c 0) ( hg (0) hgreenold (0) s) } (8) and Y low-blue = max{y y x () c ( x () c 0) ( h (0) h (0) s) }, (9) Y B B blueold where h red-old (i), h green-old (i) and h blue-old (i) are RGB histograms of the original colour image; s=mns 0 is the number of the saturated ixels in the image and usually the same value is taken for both the uer and the lower bounds, with s 0 [0.0005, 0.005]. The default s 0 is set to Using this formulation, the bounds of the new dynamic range do not deend urely on singular extreme ixels, but can be based on a reresentative set of ixels. After the new outut scaling range, q[y low, Y high ], is obtained, the transformation (19) is to be redone with the new outut scaling range, and the defect is removed of the image shown on its histogram as significant sikes at the tail ends, therefore the enhanced results show good contrast and much better colour maintenance as well(see Fig 1). The number of brightness levels which human can easily distinguish deends on many factors, such as the average local brightness. Consequently, a dislay, which avoids this effect, will normally rovide at least 100 intensity levels within its dislay range [4, 15]. Generally seaking, the number of reconstruction levels of the enhanced image in the roosed method is usually less than that of original image to rovide roer brightness scale redistribution since all ixels in each quantisation level are to be treated similarly. For an original image with 56 levels of brightness, if the number of the brightness levels is not reduced too many, no significant degradation is erceived [4, 15]. However, in some rare cases, if the original image have extremely low dynamic range with only few intensity values, the minimum brightness levels control will be alied in the roosed method by adjusting the arameters w and v, in order to ensure that the outut dynamic range is not less than 70% that of the original to avoid over contrast enhancement. The threshold of brightness level for alying the control is set to 64[, 15]. Since in some rare cases the last q of the transformation by Equation (0) is very large, a linear contrast stretch transformation is also alied in the roosed aroach to ensure full use of the outut brightness scale. Sometimes, an enhanced image serves as the inut to a machine that traces the outline of the edges, and erhas measures the shae and size of the outline. The image enhancement emhasizes salient features of the original image and simlifies the rocessing task for a data extraction machine. In this case, the roosed method can be alied directly to each of RGB channels, using the contribution from the colour and the luminance comonents for contrast enhancement. It shows clearly better erformance in some of the test images, while introducing false colour in others. III. EXPERIMENTS In the erformance evaluation, the roosed method, which works as an automatic enhancement method using arameters with default values, is comared with four classical enhancement methods (linear contrast stretching, contrast reverse, gamma correction and histogram equalization [, 15, 0] and some recent develoed histogram equalization based methods, such as DRSHE, BPDHE and GC-CHE using test images. The test images include well-known tyical test images including Mountain, Scene, Meat etc, with an image resolution of 500x36 or 71x481 or 768x768 or 731x487 ixels. The results enhanced by the roosed method with different values of its arameters are shown in Fig.1b, Fig.1c and Fig.1d. In Fig1b, the tested image Mountain was enhanced by the roosed method without outut range boundary control. In Fig1.c, the tested image Mountain was enhanced by the roosed method with t= and v=1. In Fig1.d, the tested image Mountain was enhanced by the roosed method with w =3 and v=. It is observed from the exerimental results shown in Figs. 1-4 that the roosed enhancement algorithm can effectively enhance the overall contrast and the sharness of the test images. A significant amount of details that could not be seen in the original images has been clearly revealed. For the tested colour images, better results of the comared techniques, such as linear contrast stretching, contrast reverse, gamma correction and histogram equalization, are obtained by first converting the image to the Hue, Saturation, Intensity colour sace and then alying the comared techniques to the Intensity comonent only. However, even this method does not fully maintain colour fidelity for the comared techniques [11] (see Fig.1f, Fig.3c and Fig.4c), while the roosed technique show much better colour maintenance than other techniques (see Fig.1c, Fig.1d, Fig.3b and Fig.4b). From the outut images, these test images were enhanced by histogram equalization, the histogram equalization with over-enhanced arts of the images and highlighted blocking artefacts caused by image comression (see Fig.c and Fig.4c). In terms of revealing the details in dark areas of images with a broad low histogram, the erformance of the roosed aroach is much better than the other image enhancement methods. The details in the dark areas of the test images, Mountain, Light and Scene, are much more visible than those in the original image while the colour of the images are better reserved. The linear stretching failed to make significant enhancement for these images.

6 Z. Xu et al.: Colour Image Enhancement by Virtual Histogram Aroach 709 a) ` b) c) d) e) f) Fig. 1. The enhancement results for test image Mountain, a) original image, b) outut of the roosed aroach-1, c) outut of roosed aroach-, d) outut of roosed aroach-3, e) outut of modified linear stretching, f) outut of histogram equalisation.

7 710 IEEE Transactions on Consumer Electronics, Vol. 56, No., May 010 a) b) c) d) e) f) Fig.. The enhancement results for test image Meat, a) original image, b) outut of the roosed aroach, c) outut of histogram equalisation, d) outut of linear stretching, e) outut of contrast reverser, f) outut of modified linear stretching. a) b) c) d) e) f) Fig. 3. The enhancement results for test image Light, a) original image, b) outut of the roosed aroach, c) outut of histogram equalisation, d) outut of linear stretching, e) outut of gamma correction, f) outut of GC-CHE.

8 Z. Xu et al.: Colour Image Enhancement by Virtual Histogram Aroach 711 a) b) c) d) e) ` f) Fig. 4. The enhancement results for test image Scene, a) original image, b) outut of the roosed aroach, c) outut of histogram equalisation, d) outut of gamma correction, e) outut of PBDHE, f) outut of DRSHE. Contrast reverse transfer function [] and gamma correction [, 4] are often helful in visualizing details in dark areas of an image. Therefore, the results of the reverse function and gamma correction are used in the exeriments for comarison (see Fig.e, Fig.3e and Fig.4d). The reverse function, which was clied at 10% inut amlitude levels to maintain the outut amlitude within the range of unity, was alied to the inverted original image, and then the resultant image was reverted after the enhancement rocess. The modified contrast stretch was obtained by truncated inut scale range at both the low and the high ends, corresonding to the 1% histogram distribution. IV. CONCLUSION In this aer, a new hybrid aroach based on a virtual histogram modification for colour image enhancement is roosed. The novelty of the roosed method is that colour image enhancement is based on modification of a virtual histogram distribution, which is a new way to integrate colour and brightness information extracted from salient local features, for global contrast enhancement. The secial contributions of the roosed method are the outut value scaling bounds control and outut range boundary control for the enhancement mechanism to ensure the better maintenance of colour for the enhanced images. The roosed aroach introduces the arameters to increase the visibility of secified features, ortion or asects of the image. If the arameters are set u to default values, the roosed method will work as an automatic rocess. The roosed aroach has a otential for various alications to enhance secific categories of images, such as surveillance videos/images, biomedical images and satellite images. REFERENCES [1] Cristian Munteanu and Agostinho Rosa, Gray-Scale Image Enhancement as an Automatic Process Driven by Evolution, IEEE Transactions on Systems, Man, and Cybernetics art b: Cybernetics, vol. 34, no., Aril 004. [] William K. Pratt, Digital Image Processing, John Wiley & Sons, 008. [3] G. Ramoni, N. Strobel, S. K. Mitra, and T.-H. Yu, Nonlinear Un- Shar Masking Methods for Image Contrast Enhancement, J. Electron. Imaging, vol. 5, no. 3, , [4] R. C. Gonzalez and P. Wintz. Digital Image Processing, nd Ed., Prentice Hall, 00. [5] J. B. Zimmerman, S. M. Pizer, E. V. Staab, J. R. Pery, W. McCartney, and B. Brenton, An evaluation of the effectiveness of adative histogram equalization for contrast enhancement, IEEE Trans. Med. Imag., vol. 7, , Dec [6] G. Park, H. Cho, and M. Choi, A Contrast Enhancement Method using Dynamic Range Searate Histogram Equalization, IEEE Trans. on Consumer Electronics, Vol. 54, No. 4, Nov [7] N. S. P. Kong and H. Ibrahim, Color Image Enhancement Using Brightness Preserving Dynamic Histogram Equalization, IEEE Trans. on Consumer Electronics, Vol. 54, No. 4, Nov [8] T. Kim and J. Paik, Adative Contrast Enhancement Using Gain- Controllable Clied Histogram Equalization, IEEE Trans. on Consumer Electronics, Vol. 54, No. 4, Nov [9] G. Deng, L. W. Cahill, and G. R. Tobin, The Study of Logarithmic Image Processing Model and Its Alication to Image Enhancement, IEEE Trans. on Image Processing, vol. 4. no. 4, Aril [10] C. Soong-Der and A. R. Ramli, "Contrast Enhancement Using Recursive Mean-Searate Histogram Equalization for Scalable Brightness Preservation," IEEE Trans. on Consumer Electronics, vol. 49, no. 4, , 003. [11] John P. Oakley and Hong Bu, Correction of Simle Contrast Loss in Colour Images, IEEE Transactions on Image Processing, Vol. 16, No., , February 007. [1] Wilhelm Burger and Mark J Burger, Digital Image Processing, Sringer, 008. [13] Soong-Der Chen, Abd. Rahman Ramli, Minimum Mean Brightness Error Bi-Histogram Equalization in Contrast Enhancement, IEEE Trans. on Consumer Electronics, Vol. 49, No. 4, , November 003.

9 71 IEEE Transactions on Consumer Electronics, Vol. 56, No., May 010 [14] J.A. Stark, Adative Image Contrast Enhancement Using Generalization of Histogram Equalization IEEE Trans Image Processing, vol 9, no. 5, , May 000. [15] Milan Sonka, Vaclav Hlavac and Roger Boyle, Image rocessing, Analysis, and Machine Vision, Brooks/Cole, 001. [16] S. M. Pizer et al., Adative Histogram Equalisation and its Variations, Comuter Vision, Grahics and Image Processing, 39,3, , Set [17] W. Yu, C. Qian, and Z. Baeomin, "Image Enhancement Based on Equal Area Dualistic Sub-Image Histogram Equalization Method," IEEE Transactions on Consumer Electronics, vol. 45, no. 1, , [18] E.L. Hall, Almost Uniform Distribution for Comuter Image Enhancement, IEEE Trans. Comuters, C-3,, 07-08, 974. [19] W. Frei, Image Enhancement by Histogram Equalization, Grahics and Image Processing, 6, 3, 86-64, [0] H.R. Wu and K.R. Rao, Digital Video Image Quality and Percetual Coding, CRC Press, 006. [1] D. J. Jobson, Z. Rahman, and G. A. Woodell, Proerties and erformance of a center/surround retinex, IEEE Trans. Image Process., vol. 6, no. 3, , Mar [] K. Yeong-Taeg, "Contrast Enhancement Using Brightness Preserving Bi- Histogram Equalization," IEEE Transactions on Consumer Electronics, vol. 43, no. 1,. 1-8, [3] Z. Rahman, D. J. Jobson, and G. A. Woodell, Retinex rocessing for automatic image enhancement, J. Electron. Imag., vol. 13, no. 1, , Jan [4] Tzu-Cheng Jen and Sheng-Jyh Wang, Generalized Histogram Equalization based on Local Characteristics in Proceeding of IEEE ICIP 006, BIOGRAPHIES Zhengya Xu received the Master in Information Technology from the Vrije Universiteit Brussel, Brussels Belgium in 1994, and the PhD degree from Swinburne, University of Technology, Victoria, Australia in 000. He was a research engineer with Aeronautics Research Institute, Ministry of Aviation and Aerosace Industry Beijing, China, and a senior research fellow with Monash University, Australia. He works currently as a senior research fellow on Comuter & Network Engineering at RMIT University Australia. His research interests include image and video rocessing, comuter vision, biometric attern recognition, object tracking, software and embedded systems develoment. Bin Qiu (M 93-SM 00) received the B.Eng degree from Beijing Jiaotong University, in 198, and the MSc and PhD degrees from University of Manchester, Institute of Science and Technology, UK, in 1987 and 1989 resectively. He was a lecturer at Victoria University from 1990 to He joined Monash University in 1995 as a senior lecturer, where he is currently an associate rofessor. Since 1987, he has articiated and headed research rojects in the fields of digital and comuter communication systems and networks, signal rocessing, image rocessing and the alications of neuro-fuzzy techniques in the above areas. He is also interested in the design and imlementation of digital systems. Xinghuo Yu (M'91-SM'98-F'08) received the B.Eng and M.Eng degrees from the University of Science and Technology of China, Hefei China, in 198 and 1984,and the PhD degree from South-East University, Nanjing China in 1988, resectively. He is now with RMIT University, Australia, where he is the Director of RMIT Platform Technologies Research Institute and Professor of Information Systems Engineering. Prof Yu's research interests include variable structure and nonlinear control, signal rocessing, comlex and intelligent systems. He has ublished over 300 refereed aers in technical journals, books and conference roceedings as well as coedited 10 research books. Prof Yu served as an Associate Editor of IEEE Transactions on Circuits and Systems Part I ( ) and IEEE Transactions on Industrial Informatics ( ), and is serving as an Associate Editor of IEEE Transactions on Industrial Electronics and several other scholarly journals. He has been on the rogram committees of many international conferences, and co-chaired several international conferences. Prof. Yu was the sole reciient of the 1995 Central Queensland University Vice Chancellor's Award for Research. He is a Fellow of Institution of Engineers Australia, and was made Emeritus Professor of Central Queensland University Australia in 00 for his long term significant contributions. Hong Ren Wu received the B.Eng. and M.Eng. from University of Science and Technology, Beijing (formerly Beijing University of Iron and Steel Technology), P.R. China, in 198 and 1985, resectively. He received the PhD in Electrical and Comuter Engineering from the University of Wollongong, N.S.W. Australia, in Dr Wu worked on academic staff of Chisholm Institute of Technology and then Monash University, Melbourne, Australia from Aril 1990 to January 005 last as an Associate Professor in Digital Systems. Dr Wu has been with Royal Melbourne Institute of Technology, Australia, since February 005, as Professor of Visual Communications Engineering and Disciline Head, Comuter and Network Engineering in School of Electrical and Comuter Engineering. His research interests include fast DSP algorithms, digital icture comression and quality assessment, video rocessing and enhancement, embedded DSP systems and their industrial alications. He is a co-editor of the book, Digital Video Image Quality and Percetual Coding, (CRC, 006).

10 本文献由 学霸图书馆 - 文献云下载 收集自网络, 仅供学习交流使用 学霸图书馆 ( 是一个 整合众多图书馆数据库资源, 提供一站式文献检索和下载服务 的 4 小时在线不限 IP 图书馆 图书馆致力于便利 促进学习与科研, 提供最强文献下载服务 图书馆导航 : 图书馆首页文献云下载图书馆入口外文数据库大全疑难文献辅助工具