基于多层次特征表示的场景图像分类算法

发布时间：2018-05-14 20:32

本文选题：计算机应用 + 目标属性　；参考：《吉林大学学报(工学版)》2017年06期

【摘要】：针对场景图像种类增多、场景复杂度增加和场景内容增大的趋势,本文提出了一种基于多层次特征表示的场景图像分类算法。首先采用Object Bank目标属性的高层特征表示方法,经分类器预测出该图像所属的场景主题;然后在同一场景主题内,采用基于底层特征的局部约束低秩编码方法提取图像特征;在低秩编码方法中加入局部约束正则化并采用F-范数替代核范数的优化方法,减少计算复杂度,实现对场景图像较为细致的理解。这种由高层特征和底层特征相结合的多层次特征表示方法,从对象特征的粗理解到底层细节特征的详细解析,充分利用了不同特征间层层递进和互补的关系,实验结果证明了本文算法的有效性。
[Abstract]:Aiming at the increasing of scene image types, scene complexity and scene content, a scene image classification algorithm based on multi-level feature representation is proposed in this paper. Firstly, the high-level feature representation method of the Object Bank target attribute is used to predict the scene theme of the image by classifier, and then the local constrained low-rank coding method based on the underlying feature is used to extract the image feature in the same scene theme. The local constraint regularization is added to the low-rank coding method and the optimization method of replacing kernel norm with F-norm is adopted to reduce the computational complexity and realize a more detailed understanding of scene images. This multi-level feature representation method, which combines high-level features with lower-level features, takes full advantage of the progressive and complementary relationships between different features, from the rough understanding of the object features to the detailed analysis of the underlying features. Experimental results show the effectiveness of the proposed algorithm.
【作者单位】：重庆大学自动化学院;国家电网重庆市电力公司市区供电分公司;
【基金】：国家电网公司科技项目(SGTYHT/15-JS-191) 国家自然科学基金项目(61473050)
【分类号】：TP391.41

【相似文献】