图像正交不变矩关键技术的研究

Study on the key techniques of orthogonal invariance moment for image description

寻找和构建优良的图像特征一直是模式识别领域的一个重要研究课题。不变矩由于具有平移、灰度、尺度、旋转等多畸变不变性被广泛用于图像的模式识别、图像分类、目标识别和场景分析中。该文对常见的正交不变矩，从定义构建、径向多项式选择、相互关系、评价指标等关键技术进行了对比研究，给出规律及处理方法，对寻找最优矩提供了理论和实践参考。

Seeking for and constructing excellent image features are always a research focus in the field of pattern recognition. Since the invariant moment is highly concentrated on image features that are shifting, scaling, rotation and intensity invariant, they have been extensively employed as the invariant global features of an image in pattern recognition, image classification, target identification, and scene analysis. Key techniques for their construction, choices of radial polynomial, relations between different moments and estimation parameters have been discussed in detail. Some rules and commended treatments have been provided which will be helpful in finding the optimum moment.