利用亚像素位移估计提高激光散斑图像无失真压缩效率_英文_

利用亚像素位移估计提高激光散斑图像无失真压缩效率_英文_ 第 40 卷第 4 期 红外与激光工程2011 年 4 月 Vol.40 No.4 Apr. 2011 Infrared and LasEerng ineering Improved compressionefficiency of lossless compressionof laser speckle imagesby subpixel -accuracy displacement estimation Li Donghui ,College of Computer ,ChongQing University, Chongqing 400030a, ,Chi n Abstract: The purposeo f losselss image compression is to represent ainm age signal with the bits as f ew as possible wit hout loss of any i nformation. Compression of speckle im ages has greast ign ificance to the reduction of storage requirement of measurement systems. The coder was compoof tshee d s peckle displacement estimator, the pixel predictor and thGe o lomb coding. Firstly, speckle d isplacements were estimated for each subblockN. ext, according to correlation properties of dynamic speckle, the r epdiction model was seupt and current i xpels were predicted basoend the model. Finally, prediction errors w ere coded by the Golomb coding. The i mproved algorithm was presented aand naly zed by real experiments. The proposed algorithm improved compressioneff iciency based on speckled isplacement estimation and the bilinear interpolation. The experimental results show that thime p roved coder can ac hieve a significant improvement on bit-rate. Key words: specklei mage; image compression; subixepl speckled isplacement estimation CLC number: TN919.81 Document code: A Article ID: 1007-2276(2011)04-0631-06 利用亚像素位移估计提高激光散斑图像无失真压缩效率 李东晖 ( 重庆大学 计算机学院, 重庆 400030) 摘 要, 无失真图像压缩的目的是在保证不丢 失 任何信息的情况之下 , 用尽可能少的比特 数 表 关于同志近三年现实表现材料材料类招标技术评分表图表与交易pdf视力表打印pdf用图表说话 pdf 示 图 像 。 图像压缩对数据存储量的降低 具 有重要的意义 。 所 设计 领导形象设计圆作业设计ao工艺污水处理厂设计附属工程施工组织设计清扫机器人结构设计 的激光散斑图像无 失 真编码器由激光 散 斑 位 移 估 计 、 像素预测和哥朗布 , Golomb , 编 码 所 组 成 。 首 先 , 估 计 散 斑 位 移 , 然 后 , 根 据 激 光 动 态 散斑相关函数设计预测模型 , 并以此为基础 进 行像素预测 , 最 后 , 对预测误差进行G olomb 编 码 。 提 出了一种改进的激光散斑图像压缩算法 , 并用 实验对其进行验证和 分析 定性数据统计分析pdf销售业绩分析模板建筑结构震害分析销售进度分析表京东商城竞争战略分析 。 改进的压缩算 法 通过亚像 素散斑位移估计和双线性插值以提高压 缩效 率 。 实 验 结 果 表 明 , 改进的压缩算法在压缩 性 能方面取 得了较大的提高 。 关键词, 散斑图像, 图像压缩, 亚像素散斑位移估计 收 稿 日 期 ,2010-08-11 , 修 订 日 期 ,2010-09-19 基 金 项 目 , 中 央 高 校 基 本 科 研 业 务 费(CDJZR10180011) 作 者 简 介 , 李 东 晖 (1976,) , 男 , 博 士 , 主 要 从 事 图 像 处 理 和 激 光 散 斑 测 量 方 面 的 研 究 。 Email:liiyah@163.com [14]. existing algorithms Firstly, a review of the existing algorithms is presented followed by a description of the 0 ntoduction Ir improvement. Then experiments on real speckle images In the field of optical metrology, digital speckle are presentedfollowed by conclusionsin the end. correlation method (DSCM) has been found to be an useful methodfor non-destructive testing and deformation 1 Description of the proposed coder [1-3]measurement . Recently,DSCM has shown its special 1.1 Review of the existing algorithm merits in the measurementof surface displacementsfield [14] The existing coder is structured with a speckle and displacement gradients in materials, such as being displacement estimator, a temporal predictor and the non-contact, simple optical setup and no special [15]Golomb coder , as shown in Fig.1. Speckle imagesare preparation of special requirement for the test firstly divided into equal-size and non-overlapping [4-6]environment. The principleof the methodis to compare subblocks, then speckle displacements are estimated for the two imagesof an object before andafter loading and each subblock, as shown in Fig.2, where vectors to obtain the point pairs of maximum correlation represent speckle displacements. Next, the currentpixels coefficient. However, a huge number of speckle images are predictedby using the referencepixel and prediction may be producedin experimental phaseand applications errors are coded by the Golomb coding in the end. [7-8]such asvibration measurement . For example, a sequence Speckle displacement estimations are sent to decoder of 512 speckle patterns, eachwith 512 ×512 spatial and with codestream. Decodercan reconstructoriginal image 8-bit gray-scale resolution, occupies bytes. These specklewithout lossy of any information using the previous patterns could be stored in the computer memory of speckle image (i.e., the reference imag), edecoded systems as electronic files in order to be further prediction errors and speckle displacementestimation s. processedIn . this case, storage problem may appear, especiallyif the files must be kept for a long period of time. One way to solve this storage problem is to utilize the image compression technique.The original image can not be exactly recoveredfrom lossy compression image, because it will result in an undesirable reduction in [9] measurementperformanc e. Laser speckle images can be compressed by the Fig.1 Encoding and decoding of speckle images lossless compression techniquManye. applications, such as medical imaging, image achieving, remote sensing , presentationof artwork and historical documents, require [10]. lossless compression Recent researcheshave led to many schemes and methods for lossless image [11][12][13]compression, suchLOCO_ as I, CALIC, and FELICS. Unfortunately, the schemes that have been hitherto reported seldom exploit the speckle property, itso will Fig.2 Speckle image divided into equal -size and not be able to encode speckle images with desirable nonoverlapping subblocks compression efficiency. A compression scheme that is speckle is produced by the diffuse If dynamic tailored for the speckle imageis therefore desirable. object moving in a plane with constant velocity under This paper reportsan important improvementon the Li Donghui: Improved compressieoffnic iency of lossless co mpression 第 4 期 of laser speckle images by subpixel - accuracy displacemeesntit m ation 633 * *the illumination of a Gaussian beam, as shown in Fig.3 , (x , y ) are the intensityWhere I (x , y ) and I u v 1 u v ii- Idistributions of two speckle imagesI, and are the i i-1 **赞 赞 UUy are inter preted-mean values, =x -xand =y x u u y v v as the displacement in the vertial and horizontal direction of two speckle images when the correlation coefficient distribution is maximal. Then the predictor is expressed asbelow Fig.3 Formation of speckle 赞 赞 IU(4) (r)=I(r-?) ii1- the normalized space-time correlation function of the 赞 赞 赞 Where ? U=(U, ) is the displacement estimationU x y [16]speckle intensity fluctuation be can written as follows obtainedby equation(3) . 2 2 2 22 |v| π τω r(X,τ),exp(- )exp(- |X-σvτ|)(1) 22 2However, the displacement fields obtained by ωλ R equation (3) are integer-pixel dimensions due to the Where X is the displacement of two points in the discrete aspeof ct digital images. When a subpixel observation p lane, τ is time interval, v is the velocity speckle displacement has occurrean dextra, error will be of the diffuse object, ω is the width of beam at the introducedby the prediction, as equation (4), using the object plane, λ is the wavelength of the light, R is the integer-pixel displacement estimation and will cause an distance from object plane to observation planeand σ , undesirable increasein bi t-rate.1 +R/ρ, where ρ is the wavefron-tcurvature radius of 1.2 Description of the improved algorithm the beam at the object plan e.2 1.2.1 Analysis of variance of predictionerror σ If the factor a = ?1,the peak of the 2 2 2?x /ω +σTaking I(r) as the referencep ixel of I(r), then i -1 2i 1correlation r (Xτ) doesnot disappearwith the increase- the prediction error is [12]of interval τ . In this condition, the correlation between 赞 e(r )=I (r )-I (r )=I (r )-I (r -? r)(5) 1 i 1 i-1 2 i 1 i-1 1 Itwo points I(r) and (r -σvτ) has the peak value i -1i 赞 Where ? r=r -r is the displacement estimation of I (r ). 1 2 i 1 which does not disappear, where I andIare two i i -1The varianceof e(r ) is 1 laser speckle images fetched with interval τ. Hence, the 2 2 2 δ=E[e(r) ]=E[(I(r)-I(r)) ]=1i1i12- e pixel of past fetched speckle image can be taken asthe 2 2 E[(I(r) ]+E[I(r) ]2E[I(r)I(r)]-(6) i1i12i1i12--prediction of the currentp ixel Where E [] denotes the expectation. The normalized ?赞 赞 Ir(2) (r)=I(r-? ) ii1-correlation betweenI(r ) and I(r), denoted as r?(r, τ), i1i-12赞 Where I (r) is the prediction of currentp ixel I(r), I(r) i i i -1 equals to is the reference p ixel in the last fetched speckle image r(?r, τ)= 赞 and ? ris the displacement estimation of the subblock 1 )-E[I (r )]) (I (r )-E[I(r )])] = E[(I (r i 1 i 1 i 1 2 i 1 2 --2 which currentp ixel I(r) belongst o. i2 |H (ξ)|dξ 0 乙The speckle displacementscan be extractedby the 1 (E[I (r )I (r )]-E[I (r )]E[I (r )]) (7) i 1 i-1 2 i 1 i-1 2 2 cross - of the analysis of coefficiency distribution 2|H(ξ)| dξ 乙0 [17]correlationfun ction defined asbelo w Where the amlitudep distributionof the illumin - is H(ξ)赞 赞 0 C(U, U)=xy2 m m 2* * ation light in the object plane and is |H(ξ)| dξ 乙0 [I(x,y)-I]?[I(x, y)-I]ΣΣ i u v i i-1 u v i-1 u = 1 v = 1 (3) [16] m m m m the normalization factorof the correlationfun ction. 2 * * 2 [I(x,y)-I] ? [I(x, y)-I]ΣΣΣΣ i u v i i1 u v i1 -- Thus, the cross correlation E[ I(r)I(r)] equalsto 姨姨 u = 1 v = 1u = 1 v = 1 i112i- 2 2[I(r)-I(r)]=[荦I(r)+荦I(r)]?u(13) 121202 E[I(r)I(r)]=E[I(r)]E[I(r)]+r(?r, τ)i1i-12i1i-12 |H(ξ)| dξ 乙0 This equationholds for M ×N pixels in a subimage (8) aroundr and leads to M ×N equations.The least square 赞 approximate solutionto the linear system, denoted? asu, Replacing r?(r, τ) in equation(8) with equation( 1) produces is then determinedby E[I(r)I(r)]=E[I(r)]E[I(r)]+ i1i-12i1i-12T赞TAA?u=Ab(14) 2 2 2 2 2 2 2 |v| π τω 赞exp(- )exp(- |?-σvτ|) (9) Where |H(ξ)| dξ 22 2 乙0 ωλ R ?? ??? ?? Replacing E()() in equation(6with) equation(9), [IrIr]?? i1i12- ??A= 荦I(r)+荦I(r) ?? 12 ?? 2 ??? ? δbecomes e ? ??M , N × 2 ?Ω ??2 2 2 ? ?? δ=E[I(r) ]+E[I(r) 2E[I(r)]E[I(r]-)]-??i1i12i1i12-- e ????b= 2 [I(r)I(r)] -?? 12 2 ??2 2 2 2 ? ?2 2 ? ? |v| τωπ ? ??2exp(- )exp(- |?r-σvτ|)M , N × 2 ?Ω |H(ξ)| dξ 2 22 乙0 ωλ R 1.2.3 Predictionby interpolation (10) The predictor is rewritten as below Where σvτ represents the speckle displacement at r . It 1 赞 赞 2(15) I(r)=I(r-? r) ii-1is indicated by equation ( 10) that the variance δ is a e ++赞 赞 Where r=(x,y)?Z×Z, ? r?R×R. function of |?r -σvτ | . When ? =σvτ, i.e., no error in 2赞 赞 the displacement estimatiothen, variance δ has the e subpixel accuracyof ? r, I(r -? r) isDue to the i 1 -least valueFor. obtaining asmall varianceof prediction not the pixel. To solve the problem, the prediction can displacement estimation should improved error, the be be computed as the bilinear interpolation of four in the subpixel accuracy so that the displacement adjacent pixels in the region of the reference image estimationerror will be reduced. pointed by the displacement estimatio n, as shown in 1.2.2 Subpixel displacementestimation Fig.4. [18] Subpixel displacements can be estimatedby DIS C, a spatial-gradient basedalgorith m. Taking I(r) and I(r) as the reference speckle image 12 obtained before object motion and the current speckle image obtained after object motion respectivelIy,(r ) and 1 I(r) have the relationship asbelow 2 I(r)=I(r+u)120 (11) 乙 I(r-u)=I(r) 102Fig.4 Prediction computed as thbilie n ear interpolation Where udenotes speckle displacement at r. Expand 0 赞 displacement estimation ? r can be separatedThe equations (11 ) into Taylor seriesand neglectthe second into two parts and higher order componentsyield 赞 赞 赞? r=?U-?u(16) I(r)=I(r)+荦I(r)?u1220 (12) Where 乙()=()荦()uIrIr-Ir? 2110 + + 赞 赞 赞赞 赞 赞?U=(U,U)?Z ×Z , ?u?(u,u)?((-1,-1),(1,1))xyxy Where 荦I(r) and 荦I(r) arespatia l gradientsof I(r) and 121 赞 赞 I(r) respectively. Equations(12 ) can be rearranged as? Uand ? ucan be obtained using equation (3 ) 2 Li Donghui: Improved compression eff iciency of lossless co mpression 第 4 期 635 of laser speckle images by subpixel - accuracy displacemeesntit m ation and (14) respectivelyThe. improved predictor expressed represent speckle displacement for each subblock, every in matrix operationis seven bits of which represents the displacement in the vertical and horizontal direction respectively. Hencthee , 赞 赞(r)=I(r-? )=ii-1 side-information of fourteen bits per subblock is 赞 赞 Uu I(r-?+?)=i1-equivalent to 14/(20 ×20) =0.035bits per pixel, which is 赞 赞 赞UUuI(r-?)I(r-?+(0,1)) 1- i1i1y--marginal. 赞赞[1-uu] x x"!"! 赞 赞 赞UUu I(r-?+(1,0))I(r-?+(1,1))In order to evaluate the performance of the i-1i-1y (17) improved coder, nigh speckle images each of which is fetched under differenta subpixel displacement has been赞 If ?u=(0,0), equation( 17) reducesto equation( 4). [14] and codedby the improved coder,the existing coder [19]the JPEG_LS coder(the current lossless JPEG standard) 2 Experimental results respectively.The results provided by Tab.1 indicate that evaluation of the proposed speckle coder, an For significant improvements (about 11.1%over the existing optical and analysis system are up,set as shown in Fig.5. coder and 20.8% over the JEPG_LS coder on average) The light beam emittingfrom a laser diodeof wavelength are achieved by the improved coder.As shownin column 633 nm and output powerof 10 mW is diffusely reflected 2 of Tab.1,the bit-rate obtainedby the JPEG_LS coder by a solid minute-roughness surface. The object is has no relationship with speckle displacement for the mounted on an X-Y stage and subjected to translation reason that theMED predictor of the JPEG_LS coderis a with stepping motors controlled by a microcomputer, in spatial predictor, whose prediction accuracyis determined this way, a small displacement of the object could be by the spatial correlationof specklewhich is indicatedby conducted. The minimum speckle displacementis 0.1 μm. the speckle size. The size of speckleis about 20μ m. The reflected lights proposed coder Tab.1 Evaluation of the are captured by a CCD camera to form digitalized speckle image.The CCD camera has 512×512 pixels and Displacement JPEG_LS Existing Improved Gain1 Gain2 (dx,dy) / bpp / bpp / bpp / % / % pixel size is 7.5 μm ×7.5 μm. While the surface is accurately conductedto do two-dimensionaltranslation by (0.7, 0.7) 6.2312 4.3846 4.0447 35.0895 7.7600 the controller, displaced speckle imagesare captured by (1.5, 1.5) 6.2566 5.1566 4.6950 24.9592 8.9516 an image grabber and stored into the memory of the (2.2, 2.2) 6.2555 5.6894 5.0089 19.9281 11.960 computerin bmp formatfiles. (3.0, 3.0) 6.2312 6.0491 5.2739 15.3630 12.815 (3.7, 3.7) 6.2401 6.4006 5.4310 12.9661 15.148 (4.5, 4.5) 6.2399 6.1611 5.3424 14.3832 13.288 (5.2, 5.2) 6.2432 5.8438 5.1639 17.2876 11.634 (6.0, 6.0) 6.2516 5.4827 4.9273 21.1834 10.130 Fig.5 Setupo f experiment (6.7, 6.7) 6.2500 4.8611 4.6471 25.6464 4.4022 The size of speckle image to be encoded is 375 × Average 6.2444 5.5677 4.9482 20.7567 11.1252 375 pixels and the gray-scale resolution is 8 bits. The In Tab.1, displacemen,t (dd) is speckle displacementsize of subblock is set to be 20 ×20 pixels ensuringthe xy in micron; JPEG_LS (bpp) is bit-rate obtained by the high displacement measurement accuracy in the JPEG_LS coder; Existing is bit-rate obtained by the experiment. Additional fourteen bits are used to [14]dispersion of digital speckle displacement measuremennot ise existing coder ; Improved is bit-rate obtaine dby the [J]. Chinese Optics Letters , 2006, 4(8): 453-456. JPEG_L-SImproved improved coder; Gain1 = ×100; [6] Yan Haitao, Wang Ming. Orientation of mouse u sing digital JPEG_LS speckle correlation method [ J]. Acta Optica Sinica , 2008, 28 -ImprovedExisting Gain2= ×100. Existing (3): 467-471. (in Chinese) [7] Pedrini G, OstenW, GusevM E. High-speed idgital holographic 3 Conclusion interferometry for vibration measuremen[t J] . Appl Opt, 2006, 45(15): 3456-3462. An improved algorithm used for encoding speckle [8] Li Donghui, Guo Li. Tracking specke movement b y doube llimages has been presented and analyzed by real Kaman tengJ. Chinese Optics Letters, 2006, 4(7): 338852. lfilri[]-experimentsThe. improvement of the algorithm lies in 9 Widaa J. Effects of mage compressiono n dgta speckegrams []jjiiillthe fact that speckle displacement is estimated in [J]. Optics and Lasersin Engineering, 200339(4), : 501-506. S[10] alomon D. Data Comrpessoin [M]. New York: Springer, 200 4. subpixel accuracy to obtain the reduced variance of [11] Weinberger M J, Seroussi G, Sapiro G. LOCO-I: a low prediction errors. The variance of prediction errors is complexity, context-baseldo,ss elss image compressiona lgorithm determined by the estimation accuracy of speckle [C]//IEEE on DCC, 1996: 14-0149. displacement, well as as by the object displacement. [12] Wu X, MemonN. Context-baseadd,a ptive, losselss image coding The experimental results indicate that the compression [J]. IEEE Transactions on Communications , 1997, 45 (4): gains about 20.8% over JPEG_LS and 11.1% over the 437-444. existing coder on average can be achieved by the [13] Howard P G , Vitter J S. Fast aned ffi cient lossless image compressoin[C]//IEEE on DCC, 1993, 351-360. improved coder. The high compression ratio of the [14] Li Donghui . Lossless compression o f digital speckle images improved coder may give ground for its application to based o n speckle correlation [J]. Chinese Journal of Lasers , the lossless compression of speckle images produced in 2010, 37(2): 484-487. (in Chinese) the displacement measurement. [15] Golomb S. Run -length encodings [J]. IEEE Transactions on Information Theory , 1966, 12(3): 399-401. Reference: [16] AsakuraT , Takai N. Dynamic laser speckles andt heir application [1] Francon M. Laser Speckle and Applications in Optics [M]. to velocity measurementos f diffuse object [J]. Appl Phys , Translatedb y Arsenault H H. New York: Academic Press,1 979. 1981, 25(1): 179194. - [2] Yamaguchi I. A laser-speckle strain gauge [ J]. J Phys E : Sci [17] Dai X, Chan Y C, So A C K. Digital speckle correlation Instrum , 1981, 14(11): 12-712073. method baseodn wavelet-packet noise-reduction processing [J]. [3] Feng Weiwei , Liu Meijuan , Wang Xueqin , et a l . Feature Appl Opt , 1999, 38(16): 3474-3482. extraction and rceognition of laser speckle for special material [18] Zhou P, GoodsoKn E. Subpixel displacement and d eformation surface [J ]. Infrared and Laser Engineering , 2007, 36 (2): gradient measuremenut si ng digital image/speckle correlation 186188. (in Chinese) -(DISC)[J]. Optical Engineering , 2001, 40(8): 1613-1620. [4] Li Donghui , Guo Li , Qiu Tian. Digital speckle displacement [19] Weinberger M J, Seroussi G , Sapiro G. The L OCO -I losselss measuremenbt y a t hresholding technique [C]//SPIE, 200, 6image compression al gorithm: principles and satndardization 6150: 6150591-6150597. into JPEG-LS [ J]. IEEE Transactions on Image Processing , [5] Li Donghui , Guo Li . Kalman filtering techniques for reducing 2000, 9(8): 1309-1324.