摘要
为了得到一种简单而有效的图像特征,提出了一种频域结构相似度的图像检索方法.该方法利用彩色图像的高斯变换下相位不变、梯度的结构特征和局部二值模式描述图像特征,既弥补了相位不变性对图像对比度不变的影响,又能充分利用LBP纹理特征的旋转不变性,对彩色图像有很好的检索效果.为了验证该算法的效率,在3个彩色图像库中进行了对比实验.结果表明:该方法明显优于其他算法,比GCD_HSV,HSV,LBP算法的平均检索率分别提高了27.24%,29.86%,33.48%.
In order to obtain an effective and compound feature in image retrieval, a new algorithm is proposed based on frequency structural similarity （FSIM）. This method aims to extract phase congruent （PC）, gradient magnitude of an image smoothed with Gaussian transform, and local binary pattern feature （LBP）, which not only can fill up an efficiency of the contrast invariant of phase congruent, but also can fully utilize the rotation invariant of local binary pattern. In order to test the efficiency of the proposed method, the experiments are conducted in three databases. The experimental results show that the average retrieval precision of proposed method is respectively higher 27.24%, 29.86%, 33.48% than GCD_HSV, HSV, LBP.
出处
《宁夏大学学报（自然科学版）》
CAS
北大核心
2011年第4期321-324,共4页
Journal of Ningxia University(Natural Science Edition)
基金
国家自然科学基金资助项目(60873095
61001156)
北方民族大学科学研究资助项目(2011ZMY004)
关键词
相位不变性
局部二值模式
梯度幅值
图像结构相似度
图像检索
phase congruent
local binary pattern
gradient magnitude
frequency structural similarity
image retrieval
