在图像匹配、目标识别、视频检索等应用领域中,当尺度发生变化的时候,现有灰度检测子的检测效果会明显下降。为了克服上述缺点,构建了一种尺度不变性检测子:迭代的Harris-Laplace检测子(Ite-HL-Det).首先在多尺度的情况下使用Harris角点检测子提取初始点;然后,在整个尺度空间内使用迭代法,根据规范化拉普拉斯方程的局部极值确定某局部结构的特征尺度,并使用Harris度量来最终确定兴趣点的空间位置;同时,比较全面的给出了特征尺度的有关定义和性质。实验结果表明,当存在尺度改变和旋转变化的时候,该检测子所提取的兴趣点邻域能较好地覆盖内容相同的局部图像结构。
The performance of the detector based on gray declines with scale's changing in applied fields of image matching, object recognition and video retrieval. To get over the deficiency, a detector with scale invariance, that is iterative Harris-Laplace detector (Ite-HL-Det), was constructed. Harris corner detector was used first to extract initial points in conditions of multi-scale. Then, in the whole scale space, iterative method was applied to confirm characteristic scale of certain local structure according to the local extremum of the normalized Laplacian and finally the spatial locations of interest points by Harris measure based on the definition and property of the characteristic scale proposed comprehensively. The experimental results show that neighborhoods of interest points extracted by Ite- HL-Det can well cover the same local image regions at varying scale and rotation.