Threshold Based Adaptive Power-Law Applications In Image .
International Journal of Computer Applications (0975 – 888)Volume 47– No.7, June 2012Threshold Based Adaptive Power-Law Applications inImage EnhancementT. Romen SinghSudipta RoyKh. Manglem SinghDepartment of I.T, School ofTechnology, Assam University,Silchar – 788011, Assam, IndiaDepartment of I.T, School ofTechnology, Assam University,Silchar – 788011, Assam, IndiaDepartment of Computer Sc. andEngineering, NIT, ManipurImphal -795001, Manipur, India.ABSTRACTThis paper presents a spatial domain threshold based adaptivepower-law applications (TAPLA) in image enhancementtechnique in which adaptation is carried out with local thresholds.This is an improved version of Adaptive Power-lawTransformations (APLT) [14] in which adaptation is carried outwith local means. The computational time of APLT is windowsize dependent to find local mean while the TAPLA isindependent of window-size to find local means, which are usedto determine the local threshold values. Window-size independentof computational time is due to use of integral average image asprior process to find local mean. Like APLT, TAPLA can controlthe enhancement factors such as contrast, brightness andsharpness/smoothness with a proper choice of parameters througha single function. This method can be applied on both the greyscale and color images. In the case of color images, each channelis considered separately. TAPLA outperforms better than APLTin image quality as well as in computational ontrast,Transformations, Image sharpening, Artifact, integral averageimage.1. INTRODUCTIONImage enhancement is a process of improving the quality of animage for visual perception by human beings and to make imagesmore suitable for analysis. It is an important operation in thegeneral fields of image processing and computer vision. It is alsoimportant in low level vision applications. Enhancementoperation is a task in which a set of pixel values of one image istransformed to a new set of pixel values of a new image which isvisually more pleasing and is also more suitable for analysis.Machine vision has many important applications of digital imagesthat are captured in low contrast conditions. These images oftenencounter serious problems in recognition systems. Hencecontrast and edge enhancement is an important part to solve suchproblems. How to enhance the contrast is a vital factor in imagerecognition problem, and many methods for improving the imagequality in contrast have been proposed.Contrast enhancement is important in medical image applicationsdue to the fact that visual examination of medical images isessential in the diagnosis of many diseases such as chestradiography and mammography [3], [6]. The image contrast isinherently low due to the small differences in the X-rayattenuation coefficients.Histogram equalization [1] and Adaptive Contrast EnhancementACE [12][13][30] are the well known contrast enhancementmethods for poor intensity distribution images. In addition to this,edge enhancement is also important in image recognition system.Multi scale edge enhancement using the wavelet transform [2] isa way to enhance the contrast by enhancing the edges in scalespace since edges play a fundamental role in imageunderstanding. Curvelet transformation is well-adapted torepresent images containing edges, and it works well for edgeenhancement [5]. Curvelet coefficients can be modified inorder to enhance edges in an image. Most of the methods arededicated to a single purpose. For example, contrastenhancement technique fails in edge enhancement cases whileedge enhancement technique fails in contrast enhancementcase. Power-Law Transformations [1] is also a powerfulcontrast stretching function. T.R Singh et al [14] proposeAdaptive Power-Law Transformations for image enhancementthrough contrast stretching as well as edges sharpening. It maycause over enhancement in the non-edges region sometimes.So as to solve this problem, we propose Threshold BasedAdaptive Powe-Law Applications in image enhancementwhich is an advanced version of ALPT using a thresholdvalue[15] to control the edge region.Most of the current techniques are developed for grey scaleimages. The generalization of these techniques to colorimages is not straight forward. Unlike grey scale images, thereare some factors in color images like hue, which needs to beproperly taken care of for enhancement. Hue, saturation andintensity are the attributes of color [1]. Hue is that attribute ofcolor which decides what kind of color it is, i.e., a red or anorange. One needs to improve the visual quality of an imagewithout distorting attributes of color of the image. Severalalgorithms are available for contrast enhancement for greyscale images, which change the grey values of pixelsdepending on the criteria for enhancement without taking careabout the color attributes like hue. In color imageenhancement, it must preserve the three attributes of colorimages. Sarif Kumar Naik and C. A. Murthy [9] proposed acolor image hue preserving enhancement technique where hueunaltered transformation of the image data from RGB space toother color spaces such as LHS, HSI, YIQ, HSV, etc. is done.The proposed technique TAPLA is also applicable in colorimages. For color images, TAPLA is applied channel wise.This paper is organized as follows. Section 2 describesintegral average image as a prior process to find local mean,which does not depend on local window size. Section 3describes the technique of finding threshold value by T.RSingh et al [15 ], Section 4 describes image transformationsand adaptive transformation like APLT and ACE, Session 5describes the proposed Technique TAPLA and Session 6gives the experimental results followed by conclusions inSection 7.2. INTEGRAL IMAGESAn integral image s of an input imageI is defined as theimage in which the intensity at a pixel position is equal to thesum of the intensities of all the pixels above and to the left ofthat position in the original image and it can be defined as32
International Journal of Computer Applications (0975 – 888)Volume 47– No.7, June 2012yj 1 I(x, y)xi 1S x, y s (1)Integral sum image representation has a problem of large valueof sum. If the image dimension is very large the sum value atlocations farther from origin will be very large and may not haveenough memory space to store. In order to solve this problem, thesum value is represented as average of the intensities within thatrectangular area up to the pixel position which are underconsideration.2.1.Integral Average Imageg of an input image I is defined asintegral average imageThethe image in which the intensity at a pixel position is equal to theaverage of the intensities of all the pixels above and to the left ofthat position in the original image and it can be defined asg x, y 1x yxi 1yj 1 I(x, y)(2)g ( x, y)Figure 1: Integral average image transformation sequenceTo save the image space, the integral average image is3. LOCAL THRESHOLDINGtransformed within the image itself. Ifbe the input imageitself, self transformation is done with following equations as:The binarized image b(x, y) is found asg1First raw : g 1, y y [g 1. c c g 1, y ](3)1First column: g x, 1 x [g r, 1 r g x, 1 ]𝒃 𝒙, 𝒚 (4)(8)Other points at ( x, y ) :𝟎 𝒊𝒇 𝑰 𝒙, 𝒚 𝑻(𝒙, 𝒚)𝟏 𝒐𝒕𝒉𝒆𝒓𝒘𝒊𝒔𝒆1g x, y x y [g x, y g(r, y) r y g(x, c) c x g(r, c) r c](5)Wherer x 1, c y 1, x 2. . m and y 2. . n.Using the above three equations Eqn (3-5), Eqn (2) can begimplemented to transform a grey scale input imageto anintegral average image within itself. Its sequential diagram isshown in Figure 1. This figure shows the sequential steps (a-d) ofgtransforming the input imageinto integral average imagewithin itself using Eqns (3-5). Steps (a), (b) and (c) represent thecorresponding tables. Lower values indicate the original values,whereas upper values with decimal points indicates the calculatedaverage values. In the case of color images, average images arefound for each channel R, G and B separately.2.2.Local sum and mean calculationLocal sum s(x, y) at (x, y) within a block window of size w wcan be calculated from the integral average image g with thefollowing equation ass x, y g x d, y d x d y d g x d 1, y d 1 x d 1 y d 1 g x d 1, y d x d 1 y d g x d, y d 1 x d (y d 1)](6)where d T x, y m(x, y) 1 k1 (x,y)1 (x,y) 1(9)w 12Fromlocal sumdetermined asm x, y where I(x, y) [0,1] be the intensity of a pixel at location (x,y) of the image I. In local adaptive technique, a threshold iscalculated for each pixel, based on some local statistics of theneighborhood pixels within a block of size w w. Normallylocal techniques are time consuming to process the localstatistics like local mean and standard deviation So as tominimize the computational time of local mean calculation,TR. Singh et al [15] propose an efficient way of determininglocal threshold by using Eqn.(9). They used integral sumimage in Eqn.(1) as a prior process to determine the local sumto find local mean m(x, y). But Integral sum imagerepresentation has a problem of large value of sum. To solvethis problem, we propose integral average imagerepresentation as in Eqn (2). Local sum s(x, y) and local meanm(x,y) are determined using Eqns (6) and (7) respectively.This technique of thresholding use only mean while othertechniques like Sauvola’s [18] use both mean and standarddeviation. The thresholding technique of T.R. Singh et al [15]to determine the local threshold value for local adaptation isgiven below.s(s,y)w ws( x, y), local meanm( x, y)can be(7)where (x,y) I(x,y) - m(x,y) is the local mean deviation andk1 [0,1] is a bias which can control the level of adaptationvarying threshold value. It controls the area of foreground andbackground in the binarised image resulting a convenient wayof controlling contrast area.The lower value of k1 makes the threshold value highermeaning the pixel becomes foreground and higher value of k1lowers the threshold value, the pixel becomes background.With different values of k1 , the adaptation level of thresholdvalue can be adjusted. Based on this idea, the background33
International Journal of Computer Applications (0975 – 888)Volume 47– No.7, June 2012region and foreground region take different role in the contraststretching in enhancement technique.4. IMAGE TRANSFORMATIONSAll Image enhancement simply means, transforming an image Iinto image O using f, where f is the transformation function. Thevalues of pixels in images I and O are denoted by r [0,255] ands [0,255], respectively. As said, the pixel values r and s arerelated by the expressions f(r)(10)where f maps a pixel value r into an another pixel value s. Theresults of this transformation are mapped into the grey scalerange. So, the results are mapped into the range [0, L-1], whereL 2k, k being the number of bits in the image being considered.So, for instance, for an 8-bit image the range of pixel values willbe [0, 255].4.1. ThePower-Law Transformations th power and th root curves shown in Figure 2 can begiven by the expression:s crγ(11)This transformation function is also called as Gamma Correction(GC). For various values of γ different levels of enhancementscan be obtained. GC gives overall increase or decrease inintensity to output pixel value s. If the constant values c and γare made dynamic, there will be a various level of enhancementunlike CG. T.R Singh et al [15] used this idea in adaptive powerlaw transformation technique which is described in the nextsection.4.2. Adaptive Power-Law Transformations{I,O} [0,1] are the input and output images. The AdaptivePower-law Transformations [14] is given by the followingexpression:𝑠 𝑐 1 𝑘𝑑 𝑟 𝛾(1 𝑘𝑑 )(a)(b)(c)Figure 3: Contrast Region: (a) Original image. (b) contrastregion by mean (c) contrast region by thresholding .4.3 Adaptive Contrast Enhancement (ACE)Adaptive Contrast Enhancement (ACE) [12]–[19] is a wellknown local enhancement methods. It is expressed as:(12)where s [0,1] is the transformed enhanced output pixel value ofimage O, r [0,1] is the input pixel of the image I, c, 𝛾 and k areconstants which control the brightness, contrast andsharpness/smoothness of O.𝑑 𝑟 𝑚(𝑥, 𝑦)Figure 2 : Power Law Function at Different values of𝜸 𝟎. 𝟏, 𝟎. 𝟑, 𝟎. 𝟓, 𝟏, 𝟏. 𝟕, 𝟐. 𝟓, 𝟓(13)where 𝑚 𝑥, 𝑦 is the local mean of the input image within awindow of size 𝑤 𝑤 with r as centre at (𝑥, 𝑦) of the window.If k 0, then it is equivalent to the GC and if k 0, then the threeparameters take different roles of varying different levels ofenhancement of the output image unlike GC. There will be threecontrols depending on the different parameters such as brightness,contrast and sharpness/ smoothness of the output image.Calculation of local mean is time consuming and there will beunwanted contrast stretching for highly correlated region which isshown in Figure 3.(b). In this figure the black region indicates themore contrast, where white region indicates less contrast region.It includes the region where contrast is not necessary. Thisproblem can be controlled by using a threshold vale like in Figure3(c). Based on this condition a threshold based technique isproposed for replacing the mean m(x,y) by a threshold value𝑇(𝑥, 𝑦) in Eqn.(12).s m x, y c(r m x, y )(14)where s, r are the output and input pixel value, c is a contrastgain constant and m(x, y) is local mean within a window ofsize w w . Using this constant c , all high-frequencycomponents are amplified equally. As a consequence, thosestrong high frequency components will suffer from overenhancement.5. PROPOSED TECHNIQUEThe Threshold based Adaptive Power-law Applications inimage enhancement is similar to APLT[14] as in Eqn (12)and the difference is in Eqn (13). In Eqn (13) the meanm( x, y) is replaced by a local threshold value T ( x, y)and itcan be expressed as :d r T(x, y)(15 )where 𝑇(𝑥, 𝑦) is the local threshold value of pixels within awindow of size 𝑤 𝑤 defined in Eqn.(9). r is theconcerned pixel at (𝑥, 𝑦) as centre of the window. Thus the34
International Journal of Computer Applications (0975 – 888)Volume 47– No.7, June 2012proposed system is given by the Eqns (12) and (15). The newtechnique TAPLA is equivalent to ALPT if k1 0 in Eqn.(9). LikeALPT if k 0 it is same as GC and if k 0 there are variouscontrast adjustment levels depending on the value of d with k as apositive constant. Those pixel whose values are below thethreshold 𝑇(𝑥, 𝑦), i.e. d 0 will give higher value of γ(1-kd) andlower value of c(1 kd) resulting lower transformed value of s.Those pixels whose values are greater than the threshold 𝑇(𝑥, 𝑦),i.e. d 0 will give lower value of γ(1-kd) and higher value ofc(1 kd) resulting higher transformed value of s. If k is negativeconstant, the result will be opposite resulting a smooth imageoutput. Depending on the values of γ, there will be various levelsof enhancement unlike APLT keeping the others parametersconstant. As d depends on T(x,y) and T(x,y) depends on k1, k1 alsotake a major role in this technique. Thus in this transformationtechnique, there may not be over all increase or decrease at theoutput values. The value of d varies the level of contrastadjustment with the value of . The advantage of using Eqn.(15) is that it can control the contrast region according to thevalue of k1 in Eqn. (9), whereas APLT is not there as in Figure 3.Figure 4 shows the various levels of contrast stretching of APLTand TAPLA. Not only the quality of image is enhanced, but alsoits computational time factor is reduced. Due to uses of integralaverage image, its computational time complexity is very low ascompare with other adaptive techniques and it is closed to globaltechniques.6. EXPERIMENTAL RESULTSThe proposed technique TAPLA is tested on many categories ofimages and compare with other spatial domain adaptivetechniques like ACE and APLT qualitatively as well asquantitatively. The experiment was carried out using MATLAB7.3 (R2006b) on a PC with the following configuration: Intel Core 2Duo CPU E6550,GHz 2.33 GHz, 2GB RAM, 32 bitOS(Windows Vista).6.1Image Evaluationpixels within the block (x,y), e is a very small constant addedto the denominator to avoid division by zero. Higher value ofEME denotes a higher contrast and information clarity in theimage. Using a smaller block size b b produces a moreaccurate result, but increases processing time. This measure ofenhancement employs a more structured and rational approachthan other measures to quantify contrast enhancement. It is amore reliable and accurate measure owing to the use ofcontrast entropy and other elements of visual perception. Themeasure shows considerable consistency and correctness thatis more in conjunction with the visual quality andperceptibility of the image.Tenengrad measure is based on gradient magnitudemaximization. Tenengrad value of an image I is calculatedfrom the gradient I(x, y) at each pixel location (x, y), wherethe partial derivatives are obtained by a high pass filter likeiithe Sobel operator, with the convolution kernels x and y .The gradient magnitude is given ass x, y (ix I(x, y))2 (ix I(x, y))2(17)Tenengrad criteria is then calculated asTEN x y s(x, y)2(18)The image quality in terms of sharpness and edge informationis usually considered higher if the TEN value is larger. Insmoothening case TEN value is low for more smooth image ascompare with the original image. Hence TEN value is dependon the evaluation al results can be evaluated qualitatively orquantitatively according to the purpose. Qualitative evaluationprovides a set of images for visual perception by human beings.It is highly subjective and it is not convenient to measure toquantify the enhancement for analysis. A quantitative measure isalso needed in case of parameter based algorithms to identify theoptimum enhancement point. When there is a need to comparethe results of two enhancement techniques, a quantitative measureis highly imperative. Many measures of enhancement exist, butvery few exhibit consistencies over all types of images. Manymeasures of enhancement do not show expected measurements inimages exhibiting obvious visual contrast improvement. Our newtechnique associates contrast stretching, brightness as well assharpness/smoothness and hence we choose the measure ofenhancement by entropy (EME) [23-25] for evaluating ourcontrast enhancement results and Tenengrad measure is employedfor evaluating edge information for sharpness.1.510.500246810121416-0.5Figure 4: Different levels of contrast stretching by APLTand TAPLA with local mean and local threshold value.EME is based on entropy of contrast established on thefoundation of the Michelson contrast measure [19] and useselements of human visual perception. It is expressed as𝐸𝑀𝐸 1𝑝 𝑞𝑝𝑥 1𝐼𝑚 (𝑥,𝑦)𝑞𝑦 1 𝐼 𝑥,𝑦 𝑒 𝑙𝑜𝑔𝑛1 𝐼𝑚 (𝑥,𝑦 )𝐼𝑛 𝑥,𝑦 𝑒(16)where p q is the total number of non-overlapping blocks ofmnsize b b in the image Im n such that p and q Ix (x, y)bband Iy (x, y) are the maximum and minimum grey level value of35
International Journal of Computer Applications (0975 – 888)Volume 47– No.7, June 2012Figure 5: Images used in experiment.(a)(b)(c)(d)Figure 8: Visual comparison of output result images ofBele (254 384) : (a) Original, (b) APLT at c 2, 1.5 andk -1, (c)ACE at g -0.1, (d) TAPLA at c 2, 1.5, k 1,and k1 0.1Figure 6: Visual comparison of output images of gray scene(469 346) : (a) Original, (b) APLT at c 1, 2.5 and k 2,(c)ACE at g 2, (d) TAPLA at c 1, 2.5, k 2,and k1 0.06.(a)(b)(c(dFigure 7: Visualcomparison of gray moon)) surface(256 256) :(a) Original, (b) APLT at c 1, 1.1 and k 4, (c)ACE at g 3,(d) TAPLA at c 1, 1.1 k 4, and k1 0.06.Figure 9: Visual comparison of output result of an darkimage(412 310) : (a) Original, (b) Result by APLT atc 1.5, 0.56 and k 2, (c) ACE at g 2, (d) Result byTAPLA at c 1.5, 0569, k 2,and k1 0.236
International Journal of Computer Applications (0975 – 888)Volume 47– No.7, June 2012Figure 10: Visual comparison of output result images ofButerfly(121 122) : (a) Original, (b) APLT at c 1, 0.9 andk 2, (c) ACE at g 2, (d) TAPLA at c 1, 0.9, k 2,andk1 0.1Table 1: EME Comparison of different .6072TAPLA24004300111011303990Table 2: TAN Comparison of different 076805800TAPLA200006190021400217008220Table3 : Computational Time Comparison different methodsapplied on Lena512x512 image (in sec)Window . 11 Computational time chart of different methods atdifferent window size applied on lena 512x5126.2Result analysisThe proposed technique TAPLA is tested on many categoriesof images like dark, bright and uniform luminance images ofdifferent types including medical image, scenery and aerialviews. The test result is compared with other adaptivetechniques like ACE and APLT. Figures 5 shows some ofthe tested images. Analysis is carried out through qualitativeand quantitative measure of the output images. Figure (6-10)shows the results of different techniques for visualcomparison. Figures (6-8) show the contrast stretching, edgesharpness and smoothness of gray scale images for differenttechniques respectively. Figures (9,10) show the making ofuniformed visualization of dark image and contrast as well asedge enhancement of color image respectively. Figures (1117) show the result of proposed technique TAPLA.Quantitative analysis is also employed for this technique tocompare with other techniques. Tables 1 and 2 show EME andTAN comparison with the other techniques. Higher the valueof EME and TAN there is a high quality of contrast and edgeinformation respectively. In the case of smoothing, EME andTAN are lower than the original one. In table 1 we found thatACE has higher value of EME which measure the contrast, incertain images like in Figures (8-10). But in the case of TENwhich measure edge sharpness, TAPLA has higher value forall images which are shown in Figure (6-10) and it shown intable 2.For an algorithm, computational time complexity is a majorfactor. Normally adaptive techniques are local window sizedependent and its computational complexity is 𝑶(𝒘𝟐 𝒏𝟐 )for an image of size 𝒏 𝒏 with local window size 𝒘 𝒘 .But here in this proposed technique TAPLA, it is localwindow size independent as a result of using integral averageimage to determine local mean m(x,y). Hence itscomputational time complexity is 𝑶(𝒏𝟐) which is very closedto global techniques. Figure 11 and Table 3 show thecomputational time comparison. Table 1 and 2 show thequantitative measures.From the above experimental results we observe that thisproposed technique TAPLA is outperform than the other localtechniques.37
International Journal of Computer Applications (0975 – 888)Volume 47– No.7, June 2012Figure 12: Output result images of a motor bike (512 512) byTAPLA at c 2, 1.3 and k 2,and k1 0.06Figure 14: Output result images of room (512 512) byTAPLA at c 2, .8 and k 2,and k1 0.06Figure 13: Output result of x-ray (355 479) by TAPLA atc 1.2, 1.5 and k 2,and k1 0.17. CONCLUSIONThis paper presents an improved version of adaptive power lawtransformations by replacing the local mean by local thresholdingvalue which is determined using integral average image so as todetermine local mean without depending on local window size.This new technique TAPLA outperform as compare with APLTand ACE both in image quality and computational time factor.But still this technique also has the problem of ringing edgeartifact. TAPLA is a three parameter single function imageenhancement technique. It can enhance an image throughcontrast, brightness and sharpness/smoothness with proper choiceof respective parameters. Parameter choice is very important andif choosing of parameter adapts automatically based on the inputimage type, it should be much improve in image quality.8. ACKNOWLEDGMENTSWe thank to all the authors regarding subject matter for giving usknowledge towards development of this paper in the field ofimage enhancement technique for publication to journal.Figure 15: Output result images of a ball (512 512) byTAPLA at c 2, 1.3 and k 2,and k1 0.06.38
International Journal of Computer Applications (0975 – 888)Volume 47– No.7, June 20129. REFERENCES[1] R.C. Gonzales, R. E. Woods, Digital Image Processing,2nd Edn. 2005.[2] S. Mallat, Characterization of signals for multiscaleedges, IEEE Trans. Patt. Anal., Machine Intelligence,vol. PAMI 14, pp. 710-732, 1992.[3] J.K. Kim, J.M. Park, K.S. Song and H.W.Park, Adaptive mammographic image enhancementusing first derivative and local statistics, IEEE Trans.Medical Imaging, vol. 16, iss. 5, pp. 495-502, Oct. 1997.[4] J.L. Starck, E. Cand es, and D.L. Donoho. The curvelettransform for image denoising. IEEE Transactions onImage Processing, 11(6):131{141, 2002.[5] Nagesha and G.H. Kumar, A level crossing enhancementscheme for chest radiograph images, Elsevier, Computerin Biology and Medicines, vol. 37, iss. 10, pp. 14551460, Oct 2007.[6] Kuroda, Algorithm and architecture for real timeadaptive image enhancement, SiPS 2000, pp. 171-180,2000.[7] D.C. Chang and W.R. Wu, Image contrast enhancementbased on a histogram transformation of local standarddeviation, IEEE Trans. MI, vol. 17, no. 4, pp. 518-531,Aug. 1998.[8] G. Boccignone and M. Ferraro, Multiscale contrastenhancement, Electron. Lett., vol. 37, no. 12, pp. 751752, 2001.Figure 16: Output result images of a Tample (512 512) byTAPLA at c 2, 1.3 and k 2,and k1 0.06[9] Sarif Kumar Naik and C. A. Murthy “Hue-PreservingColor Image Enhancement Without Gamut Problem”,IEEE Transactions On Image Processing, Vol. 12, No.12, December 2003.[10] R.N. Strickland, C.S. Kim and W.F. Mcdonnel, Digitalcolor image enhancement based on the saturationcomponent, Opt. Engg, vol. 26, no. 7, pp. 609-616, 1987.[11] J. S. Lee, “Digital image enhancement and noise filteringby using local statistics,” IEEE Trans. Pattern Anal.Machine Intell., vol. PAMI-2, pp.165–168, Feb. 1980.[12] T.-L. Ji, M. K. Sundareshan, and H. Roehrig, “Adaptiveimage contrast enhancement based on human visualproperties,” IEEE Trans. Med. Imag., vol. 13, pp. 573–586, Aug. 1994.[13] L. Lucchese, S. K. Mitra, and J. Mukherjee, “A newalgorithm based on saturation and desaturation in the xychromaticity diagram for enhancement and re-renditionof color images,” Proc. IEEE Int. Conf. onImageProcessing, pp. 1077–1080, 2001.[14] T.Romen Singh, O.Imocha Singh , Kh. Manglem Singh, Tejmani Sinam and Th. Rupachandra Singh “ImageEnhancement by Adaptive Power-Law Transformations”,Bahria University Journal of Information andCommunication Technology Volume 3 Issue 1 (BUJICT2010), ISSN 1999-4974.Figure 17: Output result images of a scenery (480 480) byTAPLA at c 1.2, 1.2 and k 3,and k1 0.1[15] T.Romen Singh, Sudipta Roy , O.Imocha Singh,Tejmani Sinam, Kh. Manglem Singh “A New LocalAdaptive Thresholding Technique in Binarization”, IJCSIInternational Journal of Computer Science Issues, Vol. 8,Issue 6, No 2, November 2011 ISSN (Online): 16940814.39
International Journal of Computer Applications (0975 – 888)Volume 47– No.7, June 2012[16] Bernsen, J. 1986, Dynamic thresholding of gray-levelimages. Proc. 8th Int. Conf. on Pattern Recognition, Paris, pp1251-1255.[17] W. Niblack, 1986 An IntroductionProcessing,Prince Hall Englewood Cliffs, NJ.toImage[18] J. Sauvola and M. Pietikainen, “Adaptive document imagebinarization,” Pattern Recognition 33(2), pp 255-236,2000.[19] Jinshan Tang, A contrast based image fusion technique inthe DCT domain, Digital Signal Processing 14(2004) 218226.[20] Iyad Jafar and Hao Ying , A New Method for ImageContrast Enhancement Based on Automatic Specification ofLocal Histograms, IJCSNS International Journal ofComputer Science and Network Security, VOL. 7, July 2007.[21] Bei Tang, Guillermo Sapiro ,Color Image Enhancement viaChromaticity Diffusion, IEEE TRANSACTIONS ONIMAGE PROCESSING, VOL. 10, NO. 5,May,2001.[22] Agaian, S. S, K. Panetta A. M. Grigoryan : A new measureof image enhancement. In: IASTED Int. Conf. SignalProcessing Communication, Marbella, Spain, Sep,1922,2000.[23] Agaian, S. S., K. Panetta, A. Grigoryan. : Transform basedimage enhancement with performance measure. In: IEEETransactions on Image Processing, vol. 10,No. 3, pp.367381, March,2001.[25] Silver, B., S. S. Agaian, K. A. Panetta. : Logarithmictransform coefficient Histogram matching with spatialequalization. In: SPIE Defence and Security Symposium,Mar. 2005.[26] Narasimhan K, Sudarshan C R and Nagarajan Raju, AComparison of Contrast Enhancement Techniques inPoor Illuminated Gray Level and Color Images,International Journal of Computer Applications (0975 –8887) Volume 25– No.2, July 2011.[27] Sos Agaian, Blair Silver , and Karen Panetta,, TransformCoefficient Histogram Based Image EnhancementAlgorithms using Contrast Entropy, TIP-01692-2005.[28] Konstantinos G. Derpanis,” Integral image-basedrepresentations”, Viola, P. & Jones, M. (2001). Rapidobject detection using a boosted cascade of simplefeatures. In IEEE Computer Vision and PatternRecognition (pp. I:511–518).[29] Dah-Chung Chang* and Wen-Rong Wu, Image ContrastEnhancement Based on a Histogram Transformation ofLocal Standard Deviation, IEEE TRANSACTIONS ONMEDICAL IMAGING, VOL. 17, NO. 4, AUGUST 1998.[30] Sascha D. Cvetkovic* , Johan Schirris*, Peter H.N. deWith , Locally-Adaptive Image Contrast EnhancementWithout Noise And Ringing Artifacts , 1-4244-14377/07/ 20.00 2007 IEEE.[24] Agaian, Sos S., Blair Silver, Karen A. Panetta. : TransformCoeff
in image quality as well as in computational time. Keywords Adaptive, power-law, Image enhancement, Contrast, Transformations, Image sharpening, Artifact, integral average image. 1. INTRODUCTION Image enhancement is a process of improving the quality of an image for visual perception by human beings and to make images
VIN not powered or charger disabled ISTDBY - - 1.0 µA Power On Reset Rising VIN threshold Falling VIN threshold VPOR 3.0--2.4 3.9 2.6 VIN-BAT Offset Voltage Rising threshold Falling threshold VOS-1.0--60 22 mV Over-voltage Protection Rising Threshold VOVP 6.6 6.8 7.0 V Over-voltage Protection-Threshold Hysteresis VOVPHYS - 400 - mV CHARGE CURRENT
System Management 7950 XRS Basic System Configuration Guide Page 333 System Alarm Commands config — system — thresholds — cflash-cap-alarm cflash-id rising-threshold threshold [falling-threshold threshold] interval seconds [rmon-event-type] [startup-alarm alarm-type] — no cflash-cap-alarm cflash-id — cflash-cap-warn cflash-id rising-threshold threshold [falling
Sybase Adaptive Server Enterprise 11.9.x-12.5. DOCUMENT ID: 39995-01-1250-01 LAST REVISED: May 2002 . Adaptive Server Enterprise, Adaptive Server Enterprise Monitor, Adaptive Server Enterprise Replication, Adaptive Server Everywhere, Adaptive Se
Adaptive Control, Self Tuning Regulator, System Identification, Neural Network, Neuro Control 1. Introduction The purpose of adaptive controllers is to adapt control law parameters of control law to the changes of the controlled system. Many types of adaptive controllers are known. In [1] the adaptive self-tuning LQ controller is described.
Summer Adaptive Supercross 2012 - 5TH PLACE Winter Adaptive Boardercross 2011 - GOLD Winter Adaptive Snocross 2010 - GOLD Summer Adaptive Supercross 2010 - GOLD Winter Adaptive Snocross 2009 - SILVER Summer Adaptive Supercross 2003 - 2008 Compete in Pro Snocross UNIQUE AWARDS 2014 - TEN OUTSTANDING YOUNG AMERICANS Jaycees 2014 - TOP 20 FINALIST,
Chapter Two first discusses the need for an adaptive filter. Next, it presents adap-tation laws, principles of adaptive linear FIR filters, and principles of adaptive IIR filters. Then, it conducts a survey of adaptive nonlinear filters and a survey of applica-tions of adaptive nonlinear filters. This chapter furnishes the reader with the necessary
Highlights A large thermal comfort database validated the ASHRAE 55-2017 adaptive model Adaptive comfort is driven more by exposure to indoor climate, than outdoors Air movement and clothing account for approximately 1/3 of the adaptive effect Analyses supports the applicability of adaptive standards to mixed-mode buildings Air conditioning practice should implement adaptive comfort in dynamic .
INTRODUCTION TO LAW MODULE - 3 Public Law and Private Law Classification of Law 164 Notes z define Criminal Law; z list the differences between Public and Private Law; and z discuss the role of Judges in shaping Law 12.1 MEANING AND NATURE OF PUBLIC LAW Public Law is that part of law, which governs relationship between the State
