【正文】 學(xué)2004.[18] 胡曉峰，:人民郵電出版社，2004.[19] [D]:〔學(xué)位論文〕南京:南京理工大學(xué)，2004.[20] 陳曉娟，卜樂平，李其修. 基于圖像處理的明火火災(zāi)探測研究[J]. ,19(3):611.[21] 楊枝靈, 王開等. Visual C++數(shù)字圖像獲取處理及實踐應(yīng)用[M]. 北京: 人民郵電出版社, 2003, 1.[22] 求是科技, 蘇彥華等. Visual C++數(shù)字圖像識別技術(shù)典型案例[M]. 北京: 人民郵電出版社, 2004, 8. [23] :[學(xué)位論文]上海:上海師范大學(xué)，2004 [22], 5.[24] [J].消防技術(shù)與產(chǎn)品信息，2003,10(5):65 67.[25] 羅云林，朱瑞平基于數(shù)字圖像處理的火警監(jiān)測系統(tǒng)研究[J].遼寧工程技術(shù)大學(xué)學(xué)報，2002, 10( 6) : 30 34.[26] 何斌，:人民郵電出版社，2001.[27] 蔣先剛. 基于Delphi的數(shù)字圖像處理工程軟件設(shè)計[M]. 北京: 中國水利水電出版社, 2006, 3.[28] 周澤華, 潘保昌, 鄭勝林, 趙全友, 甘艷芬. 基于多顏色模型的車牌定位方法[J]. 微計算機(jī)信息, 2007(1): 283285. [29] 薛媛. 基于視頻圖像的火災(zāi)火焰跟蹤研究[D]:〔學(xué)位論文〕西安:西安電子科技大學(xué)，2009,1.[30] ，,6(3）:5762.[31] [J],微計算機(jī)信息(測控自動化).200 4,40(9): 2122.[32] [J].,4(4A): 327330.[33] 徐瑞鑫，[J].吉林工程學(xué)院學(xué)報，2002,9:2329.[34] 杜彥蕊,.[M][35] 吳佑壽,.[M]北京:高等教育出版社,1992[36] 求是科技, 張宏林, 蔡銳. Visual C++數(shù)字圖像模式識別技術(shù)及工程實踐[M]. 北京:人民郵電出版社, 2003, 2. [37] 程鑫，王大川，[J].,14(4):239244.[38] 范維澄，:湖北科學(xué)技術(shù)出版社1994,4(1):4752.[39] 汪志兵, 崔慧娟. 一種基于紋理特征抽取的車牌定位預(yù)處理方法[J]. 計算機(jī)應(yīng)用研究, 2004(1): 255 257. [40] [D]. [碩士學(xué)位論文].長沙：國防科學(xué)技術(shù)大學(xué)，2003.[41] 陸鋒. 基于改進(jìn)的BP神經(jīng)網(wǎng)絡(luò)進(jìn)行車牌定位的研究[J]. 蘇州大學(xué)學(xué)報(工科版), 2004, 24(6): 58.[42] 劉永信，魏平，[J].內(nèi)蒙古:內(nèi)蒙古大學(xué)學(xué)報(自然科學(xué)版), 2001,32 (6):670674.[43] 王宏等譯.《計算機(jī)視覺:一種現(xiàn)代方法》[M].北京:電子工業(yè)出版社，2004.[44] [J].西北工業(yè)大學(xué)碩士學(xué)位論文，計算機(jī)應(yīng)用技術(shù).[45] 李厚君,李玉鑑. 基于 AdaBoost 的眉毛檢測與定位[J].計算機(jī)與數(shù)字工程, 2010, 38(8): 175177.[46] Koggalage, . Reducing the Number of Training Samples for Fast Support Vectorchine Classification. Neural Inform. Process. Lett. .[47] 廉小親, 陳秀新, 蘇維均. 基于紋理特征與顏色對信息的車牌定位方法[J]. 科技情報開發(fā)與經(jīng)濟(jì), 2007, 17(2): 163164.[48] 王海濤, 黃文杰, 朱永凱, 田貴云, 姬建崗. 基于聚類分析與神經(jīng)網(wǎng)絡(luò)的車牌字符識別[J]. 數(shù)據(jù)采集與處理, 2008, 23(2): 238242. [49] 葉東毅, 何蕭玲. 前饋神經(jīng)網(wǎng)絡(luò)的一個改進(jìn)的BP學(xué)習(xí)算法[J]. 福州大學(xué)學(xué)報(自然科學(xué)版), 1998, 26(2): 2224. [50] Bai Hongliang, Liu Changping. A hybrid License Plate Extraction Method Based on Edge Statistics and Morphology[C]. In: Proceedings of the 17th International Conference on Pattern Recognition (ICPR39。04), 2004(2): 831834.[51] 趙雪春, 戚飛虎. 基于彩色分割的車牌自動識別技術(shù)[J]. 上海交通大學(xué)學(xué)報, 1998, 32(10): 49.[52] 潘中杰, 譚洪舟. 模板匹配法和垂直投影法相結(jié)合的一種新的車牌字符分割方法[J]. 自動化與信息工程, 2007(2): 1213. [53] :〔學(xué)位論文〕西安:西北工業(yè)大學(xué)，2002.[54] 李波, 曾致遠(yuǎn), 付祥勝. 基于數(shù)學(xué)形態(tài)學(xué)和邊緣特征的車牌定位算法[J]. 電視技術(shù), 2005(7): 94 96. [55] [J] .計算機(jī)工程，2008，1:1416.[56] 李盛文，鮑蘇蘇. 基于PCA+AdaBoost 算法的人臉識別技術(shù)[J]. 計算機(jī)工程與應(yīng)用, 2010,46(4): 170 174. [57] Vladimir Shapiro, Georgi Gluhchev, Dimo Dimov. Towards a Multinational Car License Plate Reco gnition System [J]. Machine Vision and Applications, 2006(17): 173–183. [58] 潘崇，朱紅斌. 基于自適應(yīng)特征選擇和SVM的圖像分類的研究[J]. 計算機(jī)應(yīng)用與軟件, 2010,27(1): 244 246.[ 59] 孟祥增, [J]. 情報雜志,2005,42(9) :1415.[60] Vladimir Shapiro, Georgi Gluhchev, Dimo Dimov. Towards a Multinational Car License Plate Reco gnition System [J]. Machine Vision and Applications, 2006(17): 173183. [61] Yuntao Cui, Qian Huang. Extracting Characters of License Plates From Video Sequences [J]. Machine Vision and Applications, 1998(10): 308320. [62] 楊家輝, 土建英. 基于色彩分割與體態(tài)紋理分析的車牌定位方法[J]. 計算機(jī)與現(xiàn)代化, 2004(11):2226.[63] 郭捷, 施鵬飛. 基于顏色和紋理分析的車牌分析車牌定位方法[J]. 中國圖形圖像學(xué)報, 2002, 7(5):472476.[64] 減晶. 基于支持向量機(jī)的火災(zāi)探測系統(tǒng)研究[J].沈陽理工大學(xué)學(xué)報, 2009,1(3):5456.[65] Huang Wei, LU Xiaobo, LING Xiaojing. Wavelet Packet Based Feature Extraction and Recognition of License Plate Characters [J]. Chinese Science Bulletin, 2005, 50(2): 97100. [66] 王穎，馮志敏，[J]., 3: 1315.[67] 宋凱，[J].電腦開發(fā)與應(yīng)用. 2002,8:2529.[68] . Terbrugge, . Nijhuis, Paannen burg, . Stevens. CNNApplications in Toll Driving [J]. Journal of VLSI Signal Processing, 1999(23), 465477. [69] Felipe . Bergo Alexandre X. Falcao Paulo . Miranda Leonardo . Automatic Image[70] 劉媛珺. 雙波段野外火災(zāi)圖像識別及目標(biāo)定位方法研究[D]. [碩士學(xué)位論文]. 南京：南京航空航天大學(xué)，2009.附錄AA Spatial Constrained KMeans Approach to Image SegmentationMing Luo , YuFei Ma , HongJiang ZhangAbstractGeneral purposed color image segmentation is a challenging and important issue in image processing related , few systems successfully handle this issue within a broad diversity of this paper, we are seeking for a practical and generic solution to image segmentation. As a fast segmentation process, Kmeans based clustering is employed in feature space , in image plane, the spatial constrains are adopted into the hierarchical Kmeans clusters on each level. The two processes are carried out alternatively and iteratively. Also, an effective region merging method is proposed to handle the over experiments show the proposed approach is fast and generic, thus practical in applications.1. Introduction Color image segmentation plays a key role in many visual applications. For example, in CBIR (contentbased image retrieval)system, regionbased image retrieval attempts to overe the deficiencies of lowlevel feature based approaches, such as color histogram,color layout,and texture[1].The MPEG4 standard also advocates objectbased video pression for high pression efficiency [2]. In fact, image segmentation is a key step for image understanding, which is a natural manner to obtain highlevel semantic[3].However,due to lacking of effective and practical approach to image segmentation,these important applications undergo a slow this paper,we attempt to provide a generic intermediate segmentation of images, instead of aiming at pixelbased precision, or objectbased segmentation. Such intermediate results can be reused by all kinds of highlevel applications according to their own requirements. There existed a number of image segmentation methods in literatures. However, they are suffering from their own problems respectively. JSEG approach [4] proposes a notion of “Jimage” to measure the confidence of pixels to be boundaries or interiors of colortexture regions, and uses a region growing method to segment the image based on the Jimages. One major problem in JSEG is caused by the varying shades due to the illumination. Some stochastic model based approaches [5] have strong assumptions on feature distribution. But it is hard to prove the correctness of those assumptions, when dealing with thousands of diverse natur