Fas Single-Image Defogging 谭志明 白向晖 王炳融 東明浩 あらまし CPU GPU720 48050 fps Absrac Bad weaher condiions such as fog, haze, and dus ofen reduce he performance of oudoor cameras. In order o improve he visibiliy of surveillance and on-vehicle cameras, we propose a fas image-defogging mehod based on a dark channel prior. I firs esimaes he amospheric ligh by searching he sky area in he foggy image. Then i esimaes he ransmission map by refining a coarse map from a fine map. Finally, i produces a clear image from he foggy image by using he amospheric ligh and he ransmission map. We achieved a run speed of 50 fps @ 720 480, wih a sofware implemenaion on a cenral processing uni (CPU) wih a graphics processing uni (GPU). This fas image-defogging mehod can be used in surveillance and driving sysems in real ime. FUJITSU. 64, 5, p. 523-528 09, 2013 523
まえがき 2.5 m PM2.5 12 2 34 56 1 100 従来技術 画像霧除去モデル x I J x x 1 A x x I xj x x A J x x A 1 x 図 -1 x I x A x J x I x I xa J x x e d x d x 霧除去の方法 Tan 7 Faal 8 524 FUJITSU. 64, 5 09, 2013
霧画像 I x 明画像 J x x A I xj xxa 1x x 画 I xax を J x を 画 に霧 を する -1 Kraz 9 He 10 Tarel 11 x I He 3.0 GHz Inel Penium 4 PC600 4001 10 20 ダークチャネル処理 He I dark x min c {r,g,b} min y x I y c I c I x x 5000 75 0 90 25 255 He 開発アルゴリズム 概要 I dark x J dark x x A 1 c x I dark x J dark xa c 0 I dark x A 1 c x x 1 I dark x /A c x 0 A FUJITSU. 64, 5 09, 2013 525
図 -2 I x AA I x x 環境光の推定 A I dark x 0.1 I x 図 -3 透過マップの推定 x 図 -4 x M coarse x min c {r,g,b} min y x I y c R G B x 図 -5 1213 R G B M fine x min c {r,g,b} I x c -2-3 -5 I x J x -4 526 FUJITSU. 64, 5 09, 2013
霧画像 CPU リ GPU リ 明画像 を 画像 CPU GPU -6 a b c d -7 x x x M x min max y x M coarse ym fine x x x 1M x /A 10 0.9 画像の復元 x J I x A /max x 0 A 0 0 実装 x GPUCPU GPU GPU GPU CPU GPU 図 -6 2.53 GHz Inel i5 CPU NVIDIA GeForce 310M GPUPC 720 480 50 fps 図 -7bd むすび FUJITSU. 64, 5 09, 2013 527
100 参考文献 1 Y. Y. Schechner e al. Insan Dehazing of Images Using Polarizaion Compuer Vision and Paern Recogniion, Proceeding of he 2001 IEEE Compuer Sociey Conference Vol.1 p.325-332 2001 2 S. Shwarz e al. Blind Haze Separaion Compuer Vision and Paern Recogniion, Proceeding of he 2006 IEEE Compuer Sociey Conference Vol.2 p.1984-1991 2006 3 J. Kopf e al. Deep Phoo Model-Based Phoograph Enhancemen and Viewing SIGGRAPH Asia 2008 4 S. G. Narasimhan e al. Ineracive De Weahering of an Image Using Physical Models Workshop on Color and Phoomeric Mehods in Compuer Vision 2003 5 S. G. Narasimhan e al. Chromaic Framework for Vision in Bad Weaher Compuer Vision and Paern Recogniion, Proceeding of he 2000 IEEE Conference Vol.1 p.598-605 2000 6 S. G. Narasimhan e al. Conras Resoraion of Weaher Degraded Images The IEEE Transacions on Paern Analysis and Machine Inelligence Vol.25 No.6 p.713-724 2003 7 R. Tan Visibiliy in Bad Weaher from a Single Image Compuer Vision and Paern Recogniion Proceeding of he 2008 IEEE Conference p.1-8 2008 8 R. Faal Single Image Dehazing. Proceeding of SIGGRAPH 2008 p.1-9 2008 9 L. Kraz e al. Facorizing Scene Albedo and Deph from a Single Foggy Image Compuer Vision, Proceeding of he 2009 IEEE 12h Inernaional Conference p.1701-1708 2009 10 K. He e al. Single Image Haze Removal Using Dark Channel Prior Compuer Vision and Paern Recogniion Proceeding of he 2009 IEEE Conference p.1956-1963 2009 11 J. P. Tarel e al. Fas Visibiliy Resoraion from a Single Color or Gray LevelIimage Compuer Vision, Proceedings of he 12h IEEE Inernaional Conference p.2201-2208 2009 12 X. Lv e al. Real-ime Dehazing for Image and Video 18h Pacific Conference on Compuer Graphics and Applicaions p.62-69 2010 13 K. He e al. Guided Image Filering Compuer Vision, Proceedings of he 11h European conference p.1-14 2010 著者紹介 谭 志明 (Tan Zhiming) 王炳融 (Wang Bingrong) 白向晖 (Bai Xianghui) 東明浩 ( ひがしあきひろ ) 528 FUJITSU. 64, 5 09, 2013