电子设计工程
電子設計工程
전자설계공정
ELECTRONIC DESIGN ENGINEERING
2012年
3期
137-141
,共5页
图像检索%局部特征%K均值%图像版权
圖像檢索%跼部特徵%K均值%圖像版權
도상검색%국부특정%K균치%도상판권
image retrieval%local features%K-Means%images copyright
对图像的内容进行准确快速的描述是图像检索技术中研究的重点和难点,传统的图像特征提取方法鲁棒性较差,无法检索出修改过的图像。SIFT特征对局部特征描述能力好,同时对尺度缩放、旋转、平移、仿射变换、光照变化、剪切、降维等修改具有良好的鲁棒性,并且可以应用在多种场景下,但Lowe提出的SIFT算子的提取复杂度和匹配复杂度都非常高。为了应用SIFT对图像的描述能力及其鲁棒性,并提高效率,对SIFT的提取算法进行了修改,消除可以引起边缘响应的部分极值点,消除图像细节丰富的局部过邻近点,消除图像背景中的低对比度点,以降低算法复杂度。同时在特征提取时增加位置限制和幅度限制以降低特征点的数量,从而在匹配效率上也能得到提高。仿真实验表明,该方法在保证图像检索准确度的同时,提高了算法的效率上。
對圖像的內容進行準確快速的描述是圖像檢索技術中研究的重點和難點,傳統的圖像特徵提取方法魯棒性較差,無法檢索齣脩改過的圖像。SIFT特徵對跼部特徵描述能力好,同時對呎度縮放、鏇轉、平移、倣射變換、光照變化、剪切、降維等脩改具有良好的魯棒性,併且可以應用在多種場景下,但Lowe提齣的SIFT算子的提取複雜度和匹配複雜度都非常高。為瞭應用SIFT對圖像的描述能力及其魯棒性,併提高效率,對SIFT的提取算法進行瞭脩改,消除可以引起邊緣響應的部分極值點,消除圖像細節豐富的跼部過鄰近點,消除圖像揹景中的低對比度點,以降低算法複雜度。同時在特徵提取時增加位置限製和幅度限製以降低特徵點的數量,從而在匹配效率上也能得到提高。倣真實驗錶明,該方法在保證圖像檢索準確度的同時,提高瞭算法的效率上。
대도상적내용진행준학쾌속적묘술시도상검색기술중연구적중점화난점,전통적도상특정제취방법로봉성교차,무법검색출수개과적도상。SIFT특정대국부특정묘술능력호,동시대척도축방、선전、평이、방사변환、광조변화、전절、강유등수개구유량호적로봉성,병차가이응용재다충장경하,단Lowe제출적SIFT산자적제취복잡도화필배복잡도도비상고。위료응용SIFT대도상적묘술능력급기로봉성,병제고효솔,대SIFT적제취산법진행료수개,소제가이인기변연향응적부분겁치점,소제도상세절봉부적국부과린근점,소제도상배경중적저대비도점,이강저산법복잡도。동시재특정제취시증가위치한제화폭도한제이강저특정점적수량,종이재필배효솔상야능득도제고。방진실험표명,해방법재보증도상검색준학도적동시,제고료산법적효솔상。
Fast accurate description of image content is the focus and difficulties of image retrieval technology, and the traditional method of robust image feature extraction is poor, unable to retrieve the modified image. Characterization of the ability of local good SIFT algorithm is good at partial characterization description, while the scale scaling, rotation, translation, affine transformations, illumination changes, cut, dimension reduction and other changes, it has good robustness and can be applied in a variety of scenarios, but Lowe proposed the SIFT operator complexity of the extraction and matching complexity is very high. In order to apply SIFI" description of the image capacity and its robustness, and efficiency, this extraction of SIFY algorithm has been modified to eliminate some of the response can cause the edge of extreme points, eliminating the image detail is too close to a wealth of local points, to eliminate image in the background of low contrast points to reduce the complexity of the algorithm. Also increased position in the feature extraction limit and rate limit to reduce the number of feature points, resulting in matching efficiency can be improved. Simulation results show that the method to ensure the accuracy of image retrieval at the same time, improve the efficiency of the algorithm.