摘要
在基于内容的图像检索中,一个关键的问题是图像视觉内容的表述。而传统的颜色,形状和纹理特征对于图像内容的表述尚且不够完备。为进一步提高检索准确率,针对人眼视觉特性,该文提出了一种基于多尺度相位特征的图像检索方法。该方法首先采用尺度空间理论得到图像的多尺度描述,然后通过复数可调滤波(complex steerable filtering)提取图像的多尺度相位信息并利用直方图投影获取全局统计的多尺度相位特征。在通用数据库COREL5000上的实验结果表明,该特征相对经典的颜色特征提高至少5%检索准确率,且能对之提供有效补充。
One related key issue in Content Based Image Retrieval （CBIR） is the representation of image visual content. However, traditional image features such as color, shape and texture are not capable of representing the visual content completely. So as to improve the retrieval accuracy, an image retrieval method based on Multi-scale Phase Feature （MPF） is proposed according to the human vision. Firstly, scale space theory is adopted here to decompose the image into Multi-scale Description （MD）. And then the global statistical MPF is acquired by histogram projection from the muki-scale phase information, which is extracted by complex steerable filtering of MD. Finally, experiments on general purpose database COREL 5,000 demonstrate that the proposed MPF has a no less than 5% accuracy improvement over classic color features, and it also effectively complements classic color features.
出处
《电子与信息学报》
EI
CSCD
北大核心
2009年第5期1193-1196,共4页
Journal of Electronics & Information Technology
基金
教育部-微软重点实验室科研基金(05071802)资助课题
关键词
多尺度相位特征
尺度空间理论
复数可调滤波
图像检索
Multi-scale Phase Feature （MPF）
Scale space theory
Complex steerable filter
Image retrieval
