Abstract: Because of the diversity of different coins and the concealment of modern false coins, it has great difficulties to discriminate coins. Thus an image recognition method based on high-order statistics and local geometric characteristics is proposed. With the ratio invariance of image edge texture and image area, a variable threshold-based Robert edge detection algorithm is given. High-order features with the invariant moment, texture characteristic and regional share of the edge image, and local geometric features of different versions of the coins are selected as coin image feature vectors. The feature vectors are clustered using fuzzy C-means clustering, to realize the classification recognition of the coins. Experimental results show that the recognition rate of the proposed method can reach 98.5%, and it is adaptive to the change of environment light.

Key words: image recognition, feature extraction, high-order statistics, clustering analysis

摘要： 由于硬币具有多样性特点及现代假币手段的隐蔽性，这给硬币鉴伪带来了很大的困难，为此提出了一种基于高阶统计量与局部几何特征相结合的硬币图像识别方法。利用图像边缘纹理和图像面积的比值不变性，给出了一种变阈值的Robert边缘检测算法。将边缘图像的不变矩、纹理特征以及区域占有率等高阶统计量，以及不同版本硬币的局部几何特征量作为硬币图像的特征向量，采用模糊C均值聚类方法对其进行聚类分析，从而实现硬币的分类识别。实验结果表明该方法的识别率可以达到98.5%以上，并对环境光照的变化有很强的适应性。

关键词: 图像识别, 特征提取, 高阶统计量, 聚类分析