为了克服PC印鉴识别系统体积大、成本高并且识别精度低、适应性差的缺点, 设计了基于SIFT-SVM的嵌入式印鉴识别系统。系统计算印鉴图像的SIFT特征并进一步构造基于SIFT匹配的印鉴特征向量, 将构造的印鉴特征向量输入支持向量机进行训练, 并采用遗传算法优化SVM的惩罚因子和核参数, 使识别性能最优。系统以ARM11处理器作为印鉴特征提取和识别的核心单元, 采用ZC301摄像头进行图像采集, 并以蜂鸣器、LED、LCD作为报警显示设备。实验表明, 该系统对印鉴在模糊、旋转等多种情形下取得了93.5%的良好识别率, 且当笔划加粗仅5%时仍具有很好的适用性。系统具有识别精度好、适用性强, 体积小、设计灵活的优点。

In order to overcome the shortcomings of large volume, high cost, low recognition accuracy and poor adaptability in PC seal identification system, an embedded seal imprint identification system based on SIFT-SVM was designed. The system first computed the SIFT characteristics of the seal image and further constructed feature vector based on SIFT matching. The seal feature vector was used to train support vector machine. The SVM’s penalty factor and kernel parameter were optimized by genetic algorithm to get the best recognition performance. The system used the ARM11 processor as the core unit of the feature extraction and recognition and used the ZC301 camera to acquire image. The buzzer, LED and LCD were used as alarm display devices. The experimental results show that the system achieves a good recognition rate of 93.5% in the cases of seal blur, rotation and so on. It still has good applicability when seal’s stroke is only in 5% thickening. The system has the advantages of good recognition accuracy, strong applicability, small volume and flexible design.