基于加权优化的自然场景分类方法研究  

Classification method of natural scene based on weighted optimization

作　　者：张亚杰[1] 马秀梅[1] 张轩雄[1] 

机构地区：[1]上海理工大学信息与控制工程研究所上海市教委光电系统智能控制重点实验室,上海200093

出　　处：《计算机应用研究》2016年第5期1591-1595,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61274105);沪江基金资助项目(C14002)

摘　　要：为了提高了静态图像中自然场景分类的识别精度,采用一种基于加权优化的聚类方法。将文本领域的文字激活力矩阵方法应用到图像分类领域,将视觉词汇的数目减少使得运行时间减少,并降低了对存储内存的占用。在特征编码阶段采用局部约束线性编码算法,并将其应用在空间金字塔模型的向量量化中,该算法能有效降低量化误差改善分类效果。实验结果表明,提出的基于加权优化的局部约束线性编码算法能够获得更好的分类效果。In order to improve the classification accuracy of natural scenes in a static image,this paper used a clustering method based on weighted optimization. It applied the word activation force matrix method in text field to image classification,and reduced both the number of visual words and the running time,besides,reduced the occupation of memory. In the coding stage,the paper used the locality-constrained linear coding algorithm to vector quantization in the spatial pyramid model. The proposed algorithm can effectively reduce the quantization error to improve the classification result. The experimental results show that the proposed LLC algorithm based on weighted optimization can get a better classification effect.

关 键 词：自然场景分类 加权优化 文字激活力矩阵 局部约束线性编码 支持向量机 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]