国防科技大学学报
國防科技大學學報
국방과기대학학보
JOURNAL OF NATIONAL UNIVERSITY OF DEFENSE TECHNOLOGY
2014年
2期
129-133
,共5页
郭军%周晖%朱长仁%肖顺平
郭軍%週暉%硃長仁%肖順平
곽군%주휘%주장인%초순평
空间金字塔模型%梯度方向二进制模式%局部特征描述%图像分类
空間金字塔模型%梯度方嚮二進製模式%跼部特徵描述%圖像分類
공간금자탑모형%제도방향이진제모식%국부특정묘술%도상분류
spatial pyramid model%binary pattern of oriented gradients%local feature descriptor%image classification
空间金字塔模型由于其优势在当前图像分类中得到了广泛应用。然而,其码本生成和特征量化这两个环节具有较高的计算复杂度。为了解决这个问题,提出了一种新的局部特征表述---梯度方向二进制模式,首先对图像稠密采样得到多个子图像块,再将每个子图像块均匀划分为2×2个网格,计算每个网格的梯度直方图,然后对所有网格的梯度主方向进行二进制编码并连接为二进制串值,该二进制串值转换的十进制数即为子图像块的特征表述,最后将该特征表述嵌入到SPM模型中。在标准分类数据库上的实验结果证明了本方法在算法耗时和分类精度上均优于基于SIFT的SPM方法。
空間金字塔模型由于其優勢在噹前圖像分類中得到瞭廣汎應用。然而,其碼本生成和特徵量化這兩箇環節具有較高的計算複雜度。為瞭解決這箇問題,提齣瞭一種新的跼部特徵錶述---梯度方嚮二進製模式,首先對圖像稠密採樣得到多箇子圖像塊,再將每箇子圖像塊均勻劃分為2×2箇網格,計算每箇網格的梯度直方圖,然後對所有網格的梯度主方嚮進行二進製編碼併連接為二進製串值,該二進製串值轉換的十進製數即為子圖像塊的特徵錶述,最後將該特徵錶述嵌入到SPM模型中。在標準分類數據庫上的實驗結果證明瞭本方法在算法耗時和分類精度上均優于基于SIFT的SPM方法。
공간금자탑모형유우기우세재당전도상분류중득도료엄범응용。연이,기마본생성화특정양화저량개배절구유교고적계산복잡도。위료해결저개문제,제출료일충신적국부특정표술---제도방향이진제모식,수선대도상주밀채양득도다개자도상괴,재장매개자도상괴균균화분위2×2개망격,계산매개망격적제도직방도,연후대소유망격적제도주방향진행이진제편마병련접위이진제천치,해이진제천치전환적십진제수즉위자도상괴적특정표술,최후장해특정표술감입도SPM모형중。재표준분류수거고상적실험결과증명료본방법재산법모시화분류정도상균우우기우SIFT적SPM방법。
Recently spatial pyramid matching (SPM)with scale invariant feature transform (SIFT)descriptor has been successfully used in image classification.Unfortunately,the codebook generation and feature quantization procedures using SIFT feature have the high complexity both in time and space.To address this problem,a feature descriptor called Binary Pattern of Oriented Gradients is presented.Firstly,the input image was densely sampled and divided into small uniform image patches.Secondly,each patch was divided into 2*2 grids uniformly.For all grids the histograms of oriented gradient were computed and all dominant directions of the histograms were coded by binary coding.Then the descriptor was generated by converting the binary number to decimal number.Finally,this descriptor was combined in the spatial pyramid domain.Experiments on popular benchmark dataset demonstrate that the proposed method always significantly outperforms the popular SPMbased SIFT descriptor method both in time and classification accuracy.