基于多尺度特征融合Hessian稀疏编码的图像分类算法

发布时间：2018-04-24 02:12

本文选题：图像分类 + 特征融合　；参考：《计算机应用》2017年12期

【摘要】：针对传统稀疏编码图像分类算法提取单一类型特征,忽略图像的空间结构信息,特征编码时无法充分利用特征拓扑结构信息的问题,提出了基于多尺度特征融合Hessian稀疏编码的图像分类算法(HSC)。首先,对图像进行空间金字塔多尺度划分;其次,在各个子空间层将方向梯度直方图(HOG)和尺度不变特征转换(SIFT)进行有效的融合;然后,为了充分利用特征的拓扑结构信息,在传统稀疏编码目标函数中引入二阶Hessian能量函数作为正则项;最后,利用支持向量机(SVM)进行分类。在Scene15数据集上的实验结果表明,HSC的准确率比局部约束线性编码(LLC)高了3~5个百分点,比支持区别性字典学习(SDDL)等对比方法高了1~3个百分点;在Caltech101数据集上的耗时实验结果表明,HSC的用时比多核学习稀疏编码(MKLSC)少40%左右。所提HSC可以有效提高图像分类准确率,算法的效率也优于对比算法。
[Abstract]:To solve the problem that traditional sparse coding image classification algorithm extracts a single type of feature and ignores the spatial structure information of the image, it can not make full use of the feature topological structure information in the feature coding. An image classification algorithm based on multi-scale feature fusion Hessian sparse coding is proposed. Firstly, the image is divided into multi-scale pyramid. Secondly, the directional gradient histogram (hog) and the scale-invariant feature transform (sift) are fused effectively in each subspace layer. Then, in order to make full use of the topological structure information of the feature, The second order Hessian energy function is introduced as the regular term in the traditional sparse coding objective function. Finally, support vector machine (SVM) is used to classify. The experimental results on the Scene15 dataset show that the accuracy of HSC is 3 ~ 5 percentage points higher than that of LLC with local constraints, and 1 ~ 3 percentage points higher than that of the comparison methods such as supporting differential dictionary learning. The experimental results on the Caltech101 data set show that the time consumption of HSC is about 40% less than that of MKLSCs. The proposed HSC can effectively improve the accuracy of image classification, and the efficiency of the algorithm is better than that of the contrast algorithm.
【作者单位】：华南理工大学电子与信息学院;
【基金】：国家自然科学基金资助项目(61202292) 广东省自然科学基金资助项目(9151064101000037)~~
【分类号】：TP391.41

【相似文献】