基于分数阶矩的仿射不变特征提取

仿射不变的特征提取在目标识别和配准中起关键作用, 图像矩是提取仿射不变特征的重要方法, 高阶矩对噪声较敏感, 实际中仅有几个由整数阶矩构造的仿射不变量可用. 本文引入分数阶矩, 它由变形累次积分定义, 不仅充分利用仿射变换映直线为直线这一特性,而且能方便地消除仿射变换前后极角因子的影响. 利用分数阶矩给出了仿射不变量的构造, 传统矩构造的不变量仅是这种构造的特例. 实验结果表明低次矩构造的不变量一般有较好的抗噪性能.

Affine Invariant Feature Extraction Based on Fractional Order Moment

Extraction of affine invariant features plays a key role in pattern recognition and image registration. Moment invariant method is one of the typical and fundamental methods for the extraction of affine invariant features. However, moment invariants with high orders are sensitive to noise. Consequently, only a few low order moments can be used in practice. In this paper, fractional order moment is proposed and defined by the modified repeated integral. It not only utilizes the property of affine transform that preserves straight lines but also eliminates the non-linear transform of polar angle under affine transformation. Based on the fractional order moment, an algorithm is presented for construction of affine invariants. Invariants which are constructed by traditional moments are only special cases of the invariants constructed by fractional order moments. Experimental results show that moments of lower degree are usually more robustness to noise.