【正文】 ，使整個(gè)系統(tǒng)具備了精確搜索含有特定目標(biāo)的圖像的能力，而非進(jìn)行近似檢索。2. 由于該系統(tǒng)適合應(yīng)用于對(duì)含有特定目標(biāo)的圖像進(jìn)行檢索的場(chǎng)合，采用示例查詢的檢索方式，對(duì)檢索結(jié)果的要求明確、具體，因此省略了用戶反饋這一用于修改檢索條件的環(huán)節(jié)，操作方便。3. 使用成功匹配的特征點(diǎn)對(duì)數(shù)作為圖像相似度標(biāo)準(zhǔn)，對(duì)檢索結(jié)果根據(jù)相似度降序排列，批判標(biāo)準(zhǔn)直觀準(zhǔn)確，與人體視覺(jué)判斷相符。綜上所述，在近四個(gè)月的時(shí)間里，本人對(duì)圖像處理、CBIR系統(tǒng)和SIFT算法進(jìn)行了大量的調(diào)查、資料搜索和學(xué)習(xí)，對(duì)CBIR技術(shù)和SIFT算法的廣闊性和應(yīng)用來(lái)說(shuō)，本文僅是初步的研究和了解，實(shí)現(xiàn)的系統(tǒng)也僅停留在試驗(yàn)階段，很難具備實(shí)際的應(yīng)用能力。對(duì)此，今后的改進(jìn)方向有如下幾點(diǎn)：1. 采用多特征融合技術(shù)綜合利用除SIFT算子之外的多種圖像特征描述算子如灰度共生矩陣、Tamura 紋理特征描述算子等全局特征描述算子進(jìn)行圖像的特征提取，是系統(tǒng)在進(jìn)行精確查找的同時(shí)又可以根據(jù)用戶要求進(jìn)行近似查找，以擴(kuò)大系統(tǒng)的實(shí)用性。2. 利用SIFT算法的改進(jìn)算法，如PCASIFT、GLOH等算法進(jìn)行特征提取，以提高系統(tǒng)特征索引的生成速度或匹配速度。參考文獻(xiàn)1 David G. Lowe. Object recognition from local scaleinvariant features[R]. International Conference on Computer Vision, Corfu, Greece (September 1999), 11501157.2 David G. Lowe. Distinctive Image Features from ScaleInvariant Keypoints[J].International Journal of Computer Vision,60,2(2004),91110.3 PENTLAND A, PICARD R W, SCLAROFF S. Photobook: tools for contentbased manipulation of image databases[J]. Proc SPIE, Storage and Retrieval for Image and Video Databases, 1994, 2185(2): 3447.4 David G. Lowe. Distinctive Image Features from ScaleInvariant Keypoints. International Journal of Computer Vision, 60, 2 (2004), pp. 91110.5 Hongjiang Zhang. A Novel Region based Image Retrieval method Using Relevance Research China.6 G. Pass and R. Zabih. Histogram refinement for contentbased image retrieval,IEEE Workshop on Applications of Computer Vision, pp. 96102, 1996.7 Smith J R, Chang S F. Transform feature for texture classification and discrimination in large database. In: Proc of IEEE In39。t Conf on Image Processing. Austin, Texas, 1994.8 Chang T, Kuo J. Texture analysis and classification with treestructured wavelet Trans on Image Processing, 1993,2(4): 429441.9 趙輝. 基于點(diǎn)特征的圖像匹配算法研究[D]. 山東:山東大學(xué),碩士學(xué)位論文,200710 肖健. SIFT特征匹配算法的研究與改進(jìn)[D]. 重慶:重慶大學(xué),碩士學(xué)位論文,201211 鞠偉. 基于SIFT特征的圖像相似性檢索技術(shù)研究[D]. 江蘇:蘇州大學(xué), 碩士學(xué)位論文,201212 林傳力. 基于內(nèi)容的圖像檢索和 Sift 算法的應(yīng)用[D]. 上海:上海交通大學(xué),碩士學(xué)位論文,200813 楊恒,王慶. 一種高效的圖像局部特征匹配算法[J]. 西北工業(yè)大學(xué)學(xué)報(bào),2009,28(2):291297.14 陸軍,王潤(rùn)生. 區(qū)域結(jié)構(gòu)及多尺度紋理分析[J]. 國(guó)防科技大學(xué)學(xué)報(bào),2000,22(6):9195.15 傅蓉. 基于小波多尺度分析的彩色圖像檢索方法[J].中國(guó)圖像圖形學(xué)報(bào),2010,9(1):1326133016 皇甫堪,陳建文,樓生強(qiáng). 現(xiàn)代數(shù)字信號(hào)處理[M]. 北京:電子工業(yè)出版社,200317 (美)岡薩雷斯,伍茲. 數(shù)字圖像處理[M]. 阮秋琦,阮宇智譯. 北京:電子工業(yè)出版社,2003,15817318 茹立云,彭瀟,蘇中等. 基于內(nèi)容圖像檢索中的特征性能評(píng)價(jià)[J]. 計(jì)算機(jī)研究與發(fā)展,2003,40(11):1566157019 周文昭,夏定元,周曼麗等. 基于內(nèi)容的圖像檢索系統(tǒng)的最新進(jìn)展[J]. 計(jì)算機(jī)工程與應(yīng)用,2003,26:11211520 劉忠偉,章毓晉. 基于特征的圖象查詢和檢索系統(tǒng)[J]. 應(yīng)用基礎(chǔ)與工程科學(xué)學(xué)報(bào),2000,8(1):6977致謝本課題的研究和論文的撰寫(xiě)工作是在我的導(dǎo)師吳海濱教授的精心指導(dǎo)下完成的。借此論文完成之際，我首先要衷心感謝吳老師幾個(gè)月來(lái)不辭辛苦的指導(dǎo)和幫助，感謝吳老師給我提供的良好的研究環(huán)境和充分的學(xué)習(xí)鍛煉機(jī)會(huì)，感謝吳老師在我的論文選題、課題研究以及論文修改上傾注的心血，沒(méi)有他的幫助，本文的研究工作也就無(wú)法順利的進(jìn)行下去。吳老師嚴(yán)謹(jǐn)?shù)膶W(xué)術(shù)態(tài)度，淵博的專(zhuān)業(yè)知識(shí)，忘我的工作態(tài)度和精益求精的工作作風(fēng)給我留下了深刻的印象，也深深地感染了我，這必將使我受益終生。其次我要感謝趙棟、徐國(guó)元、占敏、朱浩若、孟賢斌等生活上的好友，四年的大學(xué)生活你們給我留下了許多難以忘懷的點(diǎn)點(diǎn)滴滴，感謝你們?cè)谏詈蛯W(xué)習(xí)上給我的幫助與歡樂(lè)。同時(shí)，也要感謝測(cè)控095班的全體同學(xué)，你們讓我的大學(xué)生活充滿了色彩斑斕的回憶，與你們?cè)谝黄鸬臅r(shí)光將永記我的心中。最后我要感謝我的父母和家人在我求學(xué)生涯中給予的關(guān)懷與資助，他們的支持與理解一直是我克服生活與學(xué)習(xí)上各種挫折與困難的不竭動(dòng)力。附錄JIRLA C++ Library for JPEG Compressed Domain Image RetrievalAbstractIn this paper we present JIRL, an open source C++ software suite that allows to perform contentbased image retrieval in the JPEG pressed domain. We provide implementations of nine retrieval algorithms representing the current stateoftheart. For each algorithm, methods for pressed domain feature extraction as well as feature parison are provided in an objectoriented framework. In addition,our software suite includes functionality for benchmarking retrieval algorithms in terms of retrieval performance and retrieval time. An example full image retrieval application is also provided to demonstrate how the library can be is made available to fellow researchers under the LGPL.Keywords Contentbased image retrieval,pressed domain retrieval, JPEG, benchmarkingINTRODUCTIONContentbased image retrieval (CBIR) has been a well researched area for over twenty years. CBIR provides a way to query a (potentially large) image collection and retrieve images of interest based on visual characteristics such as similar clour, texture, shape etc. features.Evaluation and benchmarking is recognised as one of the major difficulties in the development of puter vision and imaging algorithms, and this is especially the case for CBIR. One particular aspect of this is the rather limited availability of implementations of existing approaches which hinders appropriate parison of new approaches with the literature. While there are a few notable exceptions, such as [4], [5], there are no open implementations of most of the published CBIR algorithms.In typical CBIR systems, image features for a dataset are extracted in an offline stage, and are then cached from a database. During retrieval, corresponding features for a query image are calculated and pared to the features stored in the database. In an online retrieval scenario,features for database images are not available and hence have to be extracted during the retrieval process, requiring highly efficient methods to allow for interactive retrieval times.The vast majority of images are stored in pressed form, typically in JPEG format. Nevertheless, almost all CBIR algorithms are based on feature extraction from the(unpressed) pixel domain while presseddomain retrieval algorithms have received relatively limited attention. JPEG is a lossy image pression technique that splits the image into blocks of 88 pixels and applies the discrete cosine transformation to each block. This separates high and low frequency information, making it easier to discard visually less important information in a quantificational process. When working in the pressed domain of JPEG, image retrieval can be performed on the raw DCT coefficients,avoiding the putationally expensive inverseDCT and thus yielding a significant speedup, making this approach suitable for online retrieval.In this paper, we present JIRL, an open source C++ library for performing pressed domain CBIR of JPEG contains implementations of nine methods representing the stateoftheart of image retrieval in the pressed JPEG domain,and provides functionality for bot