Abstract: Aiming at solving the problem of the uncertainty of the local invariant features number used for image classification, based on a variety of local distribution areas analysis, this paper proposes a completeness expression of local invariant features. Local invariant features are obtained based on the local complete graph generated by adding all the weighted spatial distribution value of Gaussian in the subtraction gray value scale space. Classification experimental results using Bag of Feature(BOF) show that compared with traditional methods, this method improves classification accuracy by 5%~10%.

Key words: completeness, local invariant, Scale Invariant Feature Transform(SIFT) algorithm, image classification, Bag of Feature (BOF) model, scale space

摘要： 针对图像分类过程中局部不变特征数量不确定性问题，提出一种局部不变特征的完备性表达方法。在分析多种局部邻域分布的基础上，构造灰度的尺度差值空间。在每个差值空间，将局部空间高斯分布关系加权汇总，得到完备性描述图，进而从完备性描述图上定位兴趣点。实验采用特征袋模型进行分类，结果表明，与传统方法相比，该方法能将分类精度提高5%~10%。

关键词: 完备性, 局部不变, 尺度不变特征变换算法, 图像分类, 特征袋模型, 尺度空间

CLC Number: 

• TP391.41