采用支持向量机的指纹图像分割

提出了一种采用支持向量机分类的指纹图像分割方法。将指纹图像分块,并根据图像块的对比度特征进行初分割,以去除灰度变化较小的白背景块,对剩下的图像块提取方向偏差和频率偏差,并根据对比度、方向偏差和频率偏差三个特征分割出特征明显的前景块和背景块,采用支持向量机将经前两次分割不能判决的图像块分为前景和前景两类;采用形态学方法进行后处理以减少分割错误;为计算频率偏差,提出了一种新的频率计算方法。在FVC指纹库上对算法进行仿真实验,结果证明了方法的有效性。

Fingerprint Image Segmentation Using Support Vector Machine

A novel method for fingerprint image segmentation using support vector machine was proposed. The low gray-scale variance background blocks were segmented by the contrast. The orientation variance and frequency variance were extracted on the left blocks and the three features: contrast, orientation variance, and frequency variance were used to segment the blocks that have obvious foreground and background features. An SVM classifier was trained for the classification of the blocks that can not be decided by the first two grades segmentation. Morphology was applied as postprocessing to reduce the number of classification errors. For calculating the frequency variance, a new method was proposed to calculate the ridge frequency. The simulations were done to the proposed algorithms on FVC database, and results show that the proposed methods are effective.