Research On Seal Recognition System Based On ARM

Seal plays a vital role in social life.In order to lighten the harm brought by forged seals to the society,it is of great practical significance to carry out the automatic seal recognition system.Aiming at the low recognition rate,poor adaptability and large volume of previous systems,the seal recognition system based on ARM is studied considering from two aspects of recognition algorithm and hardware platform in this paper.Two kinds of seal recognition algorithms based on Krawtchouk moment invariants and SIFT matching features respectively were studied according to the shape feature and the local feature point of the seal image in this paper.First,the 4 order Krawtchouk moments were constructed according to the seal's structure and shape feature,and the feature vectors were formed.Then,the recognition performance was analyzed compared with the Zernike moments under the Euclidean distance.In addition,the recognition method of multiple matching feature from point of view of algebraic and geometric was proposed based on the original SIFT algorithm.The seal feature points based on SIFT were determined,and the feature vectors of seal recognition were constructed according to the SIFT feature points matching in the seal images.The recognition performance of the two methods was analyzed under the conditions of seal rotating,blurring etc.Support vector machine was used to identify seal authenticity.In order to improve the recognition performance,genetic algorithm was introduced into parameters optimization of support vector machine,so that it can identify the best classification surface.An embedded seal recognition system based on ARM was designed.The system took ARM11 processor as the core unit,carrying ZC301 camera and display alarm device.The recognition system customized the hardware driver modules of the Linux kernel,such as the camera and the USB,and finally built the software and hardware platform of the system.It used V4L2 module to acquire seal image,and realized the driving control of the buzzer,LCD and other devices through GPIO port.The cross compilation environment was established and the recognition algorithm was successfully transplanted into the ARM11 processor.In order to facilitate the operation and management,a recognition management interface for interactive communication between the ARM and the PC was designed.The experimental results show that the divergence ratio between inter class and intra class of the Krawtchouk moments under the relative Euclidean distance is 2.93 times of the Zernike moments.The Krawtchouk moments can effectively reflect the seal image characteristics.The matching feature algorithm based on SIFT can effectively solve the recognition problem of seals in various situations such as rotation and blur.It still has good applicability when seal strokes are thickened only 5%.And it can reach a good recognition rate of 93.5% under the SVM optimized by genetic algorithm.The whole identification system is flexible in design,small in space and low in power consumption.