TitleImage Enhancement for High frame-ra Author() Saito, Yauhi; Ito, Daiuke Citation Phyic Procedia (2015), 69: 265-27 Iue Date 2015 URL http://hdl.handle.net/2433/215126 2015 The Author. Publihed by El Right open acce article under the CC BY licene(http://creativecommon.org/ Type Journal Article Tetverion publiher Kyoto Univerity
Available online at www.ciencedirect.com ScienceDirect Phyic Procedia 69 (2015 ) 265 270 10 World Conference on Neutron Radiography 5-10 October 2014 Image Enhancement for High frame-rate Neutron Radiography Y. Saito* and D. Ito Reeach Reactor Intitute, Kyoto Univerity, Oaka 590-0494, Japan Abtract High frame rate neutron radiography ha been utilized to invetigate two-phae flow in a metallic duct. However, image obtained by high frame-rate neutron radiography uffered from evere tatitical noie due to it hort epoure time. In thi tudy, a patio-temporal filter wa applied to reduce the noie in the equence image obtained by high frame-rate neutron radiography. Eperiment were performed at the B4-port of the Reearch Reactor Intitute, Kyoto Univerity, which ha a thermal neutron flu of 5.0 10 7 n/cm 2. Reult were alo compared with thoe obtained at the JRR-3, the Japan Atomic Energy Intitute, with a thermal neutron flu of 1.6 10 8 n/cm 2. 2015 The Author. Publihed by Elevier B.V. Thi i an open acce article under the CC BY-NC-ND licene 2015 The Author. Publihed by Elevier B.V. (http://creativecommon.org/licene/by-nc-nd/4.0/). Selection and and peer-review peer-review under under reponibility reponibility of Paul of Scherrer Paul Scherrer IntitutIntitut. Keyword: Type your keyword here, eparated by emicolon ; 1. Introduction Flow viualization i indipenable to undertand the dynamic in thermal hydraulic. Since a non-intruive meaurement method with high patial and temporal reolution i deirable, numerou effort have already been made to develop a method for meauring flow with better ignal quality. Neutron radiography i one of the radiographic technique that make ue of the difference in the attenuation characteritic of neutron in variou material, whoe characteritic complementary to thoe of X-ray radiography ha many advantage for fluid flow in a metallic duct [1]. Thermal neutron eaily penetrate mot dene metal, wherea they are attenuated by uch * Correponding author. Tel.: +81-72-451-2374; fa: +81-72-451-2364. E-mail addre: yaito@rri.kyoto-u.ac.jp 1875-3892 2015 The Author. Publihed by Elevier B.V. Thi i an open acce article under the CC BY-NC-ND licene (http://creativecommon.org/licene/by-nc-nd/4.0/). Selection and peer-review under reponibility of Paul Scherrer Intitut doi:10.1016/j.phpro.2015.07.037
266 Y. Saito and D. Ito / Phyic Procedia 69 ( 2015 ) 265 270 material a hydrogen, water, boron, gadolinium, and cadmium. Therefore, neutron radiography i more uitable for viualizing and meauring a multiphae flow in a metallic duct or a liquid metal flow than -ray [2]. However, the temporal limit of neutron radiography depend on the neutron flu emanated from a neutron ource. The hort epoure in neutron radiography caue evere noie due to the tatitical variation of the neutron flu [3]. Generally, time averaging can be applied to reduce uch tatitical noie from the original image. However, uch time averaging cannot be ued for motion picture, which may caue image blurring in the filtered image. Spatial filter can be ued for high frame-rate neutron radiography, however, if the S/N ratio (ignal to noie ratio) i really poor, the patial noie cannot be removed jut by applying uch patial filter. Thu, it i indipenable to reduce noie from an original image by applying a temporal and patial filter at the ame time. The purpoe of thi tudy i to apply uch a patio-temporal filter to high frame-rate neutron radiography to enhance the image quality and to compare the effect of neutron flu number from the different neutron ource. Nomenclature B E F g p t R Σ φ th δ gray level in the filtered image Statitical error Neutron fluence Gauian function poition poition epoure time patial reolution macrocopic cro ection thermal neutron flu Object thickne 2. Eperiment 2.1. Imaging Sytem The block diagram of the imaging ytem for high frame-rate neutron radiography i hown in Fig.1. The eperiment were performed by uing the B4-port of the Reearch Reactor Intitute, Kyoto Univerity [4]. A highpeed video camera (MotionPro Y4 Lite, IDT Co.Ltd.) wa teted to achieve high patial reolution (1014 1014 piel). A high-gain, fat gated image intenifier (Single MCP (GaAP)+Booter) by HAMAMATSU wa employed to amplify the light intenity of phophorecent image from a fluorecent converter ( 6 LiF/ZnS:Ag)[5]. The epoure time can be adjuted by a pule generator connected to the camera ync out. The light hielding bo equipped with no optical mirror to reduce light decay and the optical ai wa inclined at 45 to the neutron beam to avoid radiation damage. (a) Detail of imaging ytem (b) light hielding bo Fig. 1. (a) Detail of imaging ytem; (b) layout of the light hielding bo
Y. Saito and D. Ito / Phyic Procedia 69 ( 2015 ) 265 270 267 2.2. Eperimental etup The eperimental apparatu conit of a mall tainle teel tube, a tube pump and a reervoir tank. Three different diameter tube (1.0, 2.0, and 4.0mm) were teted. Air-water two-phae miture wa pumped by the tube pump and recorded by the above mentioned imaging ytem. 3. Image degradation in neutron radiography Generally image degradation in neutron radiography can be influenced by a number of ytem component here are pecific eample [6]. a) Object cattering degradation b) Geometric unharpne due to divergence of neutron beam c) Motion unharpne due to object motion d) Statitical noie due to hort epoure time The firt two contribution a) and b) hould be taken into account for tatic and dynamic neutron radiography. Lat two contribution c) and d) hould be conidered, in particular, for high frame-rate neutron radiography, where rapid fluid motion hould be oberved. Since the meaurement of neutron generated by a random proce i affected variation, the meaurement error increae with recording peed. A the meaurement error i uually given by the tandard deviation, the relative meaurement error E i given a a function of the neutron fluence F by the equation [1]: E F F (1) The neutron fluence per piel area in the image enor i given by 2 ( Σδ ) R t F φth ep (2) Where R i the patial reolution of the image, namely the real cale projected in a piel, and t i the meauring time or epoure time. Meauring time hould be adjuted depending on the fluid flow condition, where the tatitical noie can increae. The temporal reolution of dynamic neutron radiography i determined by both the light decay characteritic of the converter and the meaurement error of neutron caued by the tatitical fluctuation of neutron flu. However, both of the temporal and patial reolution would be inufficient for practical application to the rapid two-phae flow phenomena. Therefore, noiy image with hort epoure time hould be enhanced by applying ome filtering method. 4. Spatial and temporal filter In general, the value of the filtered image at a given location i a function of the value of the input image in a mall neighborhood of the ame location. Gauian low-pa filtering compute a weighted average of piel value in the neighborhood, in which the weight decreae with ditance from the neighborhood center [7]. Spatial filter and temporal filter are widely ued in image proceing and ignal proceing. In particular, Gauian low-pa filtering compute a weighted average of piel value in the neighborhood, in which, the weight decreae with ditance from the neighborhood center. However, the aumption of low patial variation in uch filtering fail at edge, which are conequently blurred by low-pa filtering. Many effort have been devoted to reducing thi undeired effect. Aniotropic diffuion [8] i a popular filtering method and the diffuion method average over etended region by olving partial differential equation, and are therefore inherently iterative. Iteration may raie iue of tability and depend on the computational architecture. On the other hand, bilateral filtering moothe
268 Y. Saito and D. Ito / Phyic Procedia 69 ( 2015 ) 265 270 image while preerving the edge, it i none iterative, local and imple [9]. The filtered image can be calculated uing the following equation: B () p N g p N ( p, σ d ) g I( p) I( ), σ i g( p, σ ) g I( p) I( ), d ( ) I( ) ( σ ) i (3) g 1 2 (, σ ) ep( 2 2σ ) N (4) σ 2π Where, I i the original gray level, B the filtered gray level, repectively. p(p,p y ) and (, y ) are poition. In Eq.(3), two Gauian function appear and the firt denote a imilarity function in the ditance, and the econd denote a imilarity function in the gray level. Bennett [10] propoed a modification of the bilateral filter for motion video image to take the motion in the image into account. At the preent tudy, we propoe a imilar imple modification to the original bilateral filter a follow: B N ( t) y [ k, + k] [ k + k] y, y g( ( p, t ) ( ) ) ( ( ) ( ) ) ( ) tp p N p, t, σ d g I p, t p I, t, σ i I, t g( ( p, t ) ( ) ) ( ( ) ( ) ) tp p N p, t, σ d g I p, t p I, t, σ i [ k, + k], (6) y [ k + k] y, y (5) (7) [ t m t m] t p, + (8) Where t p i the time domain for moothing in the time, which wa aumed to have the ame weight in the patial domain. If we input zero to the time domain m, we can obtain the original equation of the bilateral filter a written in Eq.(3). 5. Reult The propoed patio-temporal filtering method (Eq.(5)) ha been applied to image obtained at the B4-port of KUR and at the JRR-3 of JAEA. Figure 2 how the original and filtered image of two-phae miture flowing image in a mall diameter tube obtained at the B-4 port of the KUR. The epoure time wa 5m, and the ucceive image are hown at every 20m. A hown in thee image, it i recognized that the evere noie could be removed from the original image by applying the modified bilateral filter, while preerving the edge information.
Y. Saito and D. Ito / Phyic Procedia 69 (2015) 265 270 Fig. 2. (a) Original two-phae flow image (b) Filtered image.( Epoure time:5m, tube diameter:1mm) Figure 3 how the filtered and original proceed image obtained at the JRR-3M of the JAERI. The epoure time wa 2m, and the ucceive image are hown at every 60m. A ingle water droplet evaporate in a heated motel metal pool. A hown in thee image, evere ignal noie in the darker region around the evaporating bubble due to the neutron attenuation in the liquid phae. Such evere ignal noiy image can be alo enhanced by applying the propoed filtering method. Fig. 3. Water droplet evaporating in a liquid metal (a) Original image (b) Filtered image.( Epoure time:2m) 269
270 Y. Saito and D. Ito / Phyic Procedia 69 ( 2015 ) 265 270 6. Concluion Spatio-temporal filter wa propoed for high frame-rate neutron radiography by modifying the bilateral filter [6]. Eperiment were performed at the B4-port of the Reearch Reactor Intitute, Kyoto Univerity, which ha a thermal neutron flu of 5.0 10 7 n/cm 2 and at the JRR-3M, the Japan Atomic Energy Intitute for thermal neutron with a flu of 1.6 10 8 n/cm 2. It i confirmed that the noiy image can be clearly enhanced by applying a modified bilateral filter a epreed by Eq.(6) Acknowledgement We are grateful to the reactor group member of the Reearch Reactor Intitute, Kyoto Univerity for the preparation and the operation in the neutron radiography facility. A part of thi tudy i the reult of Improvement of Utilization on KUR and HL in KURRI carried out under the Strategic Promotion Program for Baic Nuclear Reearch by the Minitry of Education, Culture, Sport, Science and Technology of Japan. Reference [1] Hibiki T, Mihima K, and Matubayahi M, Application of High Frame-Rate Neutron Radiography with a Steady Thermal Neutron Beam to Two-Phae Flow Meaurement in a Metallic Rectangular Dcut, Nuclear Technology, 1995; 110: 422-435 [2] Saito Y, Mihima K, Tobita Y, Suzuki T, Matubayahi M, Lim I.C, and Cha J.E, Application of high frame-rate neutron radiography to liquid-metal two-phae flow reearch, Nuclear Intrument and Method in Phyic Reearch Section A, 2005; 542: 168-174. [3] Hibiki T, Mihima K, and Matubayahi M, Feaibility of High Frame-Rate Neutron Radiography by uing a Steady Thermal Neutron Beam with 106 n/(cm2), Nuclear Intrument and Method in Phyic Reearch Section A, 1996; 369: 186-194 [4] Saito Y, Sekimoto S, Hino M, Kawabata Y, Development of Neutron Radiography Facility for Boiling Two-Phae Flow Eperiment in Kyoto Univerity Reearch Reactor,Nuclear Intrument and Method in Phyic Reearch, 2011; A651:36-41 [5] Saito, Y., Mihima, K., Matubayahi, M., Lim, I.C., Cha, J.E., and Sim, C.H., 2005, Development of high-reolution high frame-rate neutron radiography, Nuclear Intrument and Method in Phyic Reearch Section A, 542, 309-315 [6] Harm A.A, and Wyman D.R., Mathematic and Phyic of Neutron Radiography, D. Reidel Publihing Company;1986 [7] Lee J.S, Digital image enhancement and noie filtering by ue of local tatitic, IEEE Tran., ASSP-27(1);1980: 13-18 [8] Perona P, and Malik J, Scale-pace and edge detection uing aniotropic diffuion, IEEE Tran, PAMI-12(7);1990:629-639 [9] Tomai C, and Manduchi R, Bilateral Filtering for Gray and Color Image, Proc. 6th Conf. Computer Viion; 1998 [10] Bennett E.P, Computational Video Enhancement, PhD thei, Univerity of North Carolina, the Department of Computer Scicence; 2007