不变矩稳定性及基于多级特征模型的目标识别研究

从成像过程和实际应用角度研究影响目标不变矩稳定性的各种因素，包括有限观测区域、高斯模糊和离散化处理以及在这些因素影响下不变矩的变化规律和误差．理论分析和实验还证明各不变矩自身的稳定特性不仅与阶数而且与其次数有关．这些研究和分析表明建立多尺度目标特征模型的必要性，可为不变矩在实际成像识别系统中正确有效地应用提供重要的理论和实验依据．

Moment Invariant Stability and Its Features for Object Recognition

Based on the image processing, a study is made on various factors that influence the stability of moment invariants, including the limited observation region, Gaussian blurring and discrete processing. The variations and errors of the moment invariants arising from the factors are analyzed. Theoretical analysis and experiment show that the stability of the moment invariant itself is affected not only by its order but also by its polynomial degree. The work so far done has also proved the necessity for the development of a multi-scale feature model that will provide important theoretical and experimental basis for correct and efficient application of moment invariants in the actual imaging recognition systems.