Embedded automatic fingerprint recognition system based on DSP56311 and MCS-51

Introduction
At present, biometric technology is developing rapidly, especially fingerprint recognition technology, which is known as one of the three major biometric technologies, is becoming more and more advanced. Applying automatic fingerprint recognition technology to a variety of systems that need authentication or embedding it in a large number of existing security systems will greatly improve the security of the system. In this paper, a lot of research and improvement have been made on the current automatic fingerprint recognition algorithm, which not only reduces the misrecognition rate and rejection rate of fingerprint recognition to very low (about 0.005%), but also greatly improves the speed of fingerprint feature matching. This paper describes the technical principle and key points of the embedded automatic fingerprint recognition system which uses Motorola DSP56311 and MCS-51 to design a master-slave structure.
System Diagram
The embedded automatic fingerprint recognition system has two operating modes: master mode and slave mode. In the master mode, the system runs independently, the user fingerprint recognition algorithm uses a 1:N matching method, and the fingerprint data and operation record data are stored in Flash. MCS-51 implements a simple embedded database management system that provides operations such as adding, deleting, modifying user fingerprints, querying, importing and exporting operation records, and emptying the database. In slave mode, the system can be used as the peripheral of other systems. The system does not use local fingerprint library. The fingerprint automatic recognition algorithm uses 1:1 matching method, and the reference fingerprint data comes from the external system. The hardware component diagram is shown in Figure 1.
There are two key points in designing the system: First, the hardware and software design of the communication interface between DSP56311 and MCS-51 single-chip computer. This part must ensure that the system can exchange data efficiently and quickly when making fingerprint signature search matching. Second, the selection and software implementation of fingerprint automatic recognition algorithm in DSP56311 directly affect the matching speed and effect of the system.

HI08 module configuration of interface between DSP56311 and MCS-51
The HI08 (host interface) interface is divided into two parts: the interface with the main processor and the interface with DSP. Eight registers in the interface part with the DSP kernel are mapped directly to the internal X data memory. You can use software query, interrupt driver or DMA to shake hands with the DSP kernel. Registers that interface to the main processor are mapped to eight consecutive locations in the external host bus address space, acting as an I/O mapping peripheral for the main processor.
The working mode, port properties and configuration information of HI08 interface module, as well as the data transmission mode between the DSP kernel and MCS-51, are implemented in bootstrap according to the setting of HPCR registers. The definition of HPCR is shown in Table 1. High bytes mainly set HI08 interface signal polarity and external bus type, while low bytes mainly set enabling flags related to port signal.
The data transmission of the system uses interrupt mode, so the interrupt allowed flags HCR[HTIE], HCR[HRIE], ICR[TREQ], ICR[RREQ] on both sides of HI08 and DSP kernel and MCS-51 should be set to 1 respectively. In this system, MCS-51 sends various commands or feature template data needed for fingerprint matching to DSP56311. DSP56311 returns processing results of various commands or fingerprint signature data to MCS-51 single-chip computer.
The process of sending data from MCS-51 to DSP56311 is as follows: interrupt first, set ICR[TREQ]= 1, HCR[HRIE]=1; The data is then written to the register TXH:TXM:TXL of the HI08 interface, and the ISR [TXDE] clears automatically; When HSR [HRDF]=0, the data in the register TXH:TXM:TXL will be transferred to the data receiving register HRX, and the ISR [TXDE] and HSR [HRDF] will be automatically placed at 1; ISR [TXDE] and HSR [HRDF] both produce two interrupt signals for 1, that is, the interrupt request signal for the data received by the DSP and the interrupt signal for the data continued to be transmitted by the single-chip computer (corresponding to the HRRQ pin signal).
Similarly, the process of sending data to MCS-51 by DSP56311 is that the kernel is first interrupted, HCR[HTIE]= 1, ICR[RREQ]=1; The data is then written to the HTX register and the HSR [HTDE] is automatically zeroed out; When ISR [RXDF]=0, the data in HRX will be automatically written to RXH:RXM:RXL, and both the HSR [HTDE] bit and the ISR [RXDF] bit will be automatically set to 1; Both HSR [HTDE] and ISR [RXDF] will produce two interrupt signals for 1, that is, the interrupt signal for the data transmitted by the kernel and the interrupt signal for the data received by MCS-51 (corresponding to the pin signal of HTRQ); The single-chip computer detects that the interrupt signal will read data from the register RXH:RXM:RXL of HI08 interface, and the ISR[RXDF] will be cleared automatically after reading out the data.
The connection diagram of DSP56311 to MCS-51 through HI08 interface is shown in Figure 2.

Selection and improvement of algorithm
In the two working modes, the automatic fingerprint recognition algorithm used by the system has different emphasis. In the master mode, fingerprint database has a small capacity and often requires fast and accurate matching. The focus of the algorithm is to calculate the detailed feature point data of the fingerprint image. The capacity of fingerprint database is usually large under controlled mode, so the calculation of global feature data for fingerprint classification retrieval is very important. The flow of fingerprint processing algorithms in the two working modes is basically the same, as shown in Figure 3.
The fingerprint image enhancement algorithm uses Gabor filter to process poor quality fingerprint images, which makes the rejection rate of the system very low. The binarization algorithm uses a dynamic threshold binarization algorithm based on the directional graph, which can effectively reduce the generation of pseudo-feature points and the degradation and loss of real feature points.
The fingerprint image thinning algorithm uses the classic Hildch image thinning algorithm, which meets the requirements of convergence, connectivity, topology, preservation, thinning, axis and speed. There are two kinds of signature extraction algorithms: global feature point extraction algorithm and detailed feature point extraction algorithm. The global feature point refers to the singular point of the fingerprint image, that is, the center point and the triangle point. To reduce computational effort, the detail feature point extraction algorithm only extracts two key points: the end of the ridge line (endpoint) and the branch point of the ridge line. These two feature points are often used in embedded applications to identify fingerprints very well. Fingerprint signature matching is to calculate the degree of matching between the two feature points. The global feature point calculation method uses an improved Poincare Index value calculation algorithm with good anti-jamming performance. Fingerprint detail extraction algorithm based on ridge line tracking is used in detail point calculation. Fingerprint signature matching is a method proposed by D.K.Isenor et al. to match two fingerprint images by graph matching and an algorithm based on structure matching by Andrew K.Hrechak et al. to identify fingerprints by structure matching. A structure matching algorithm based on dynamic global features is presented. The algorithm has been verified in MATLAB and the matching result is very good.
The idea of this matching algorithm is to divide the fingerprint feature point distribution map into several triangles (primitives) which are independent of each other by using global and detail feature points and some criterion. Each triangle vertex is an endpoint or a branch point, and the tangent direction at the endpoint (fingerprint extension direction) and the three branch directions of the branch point are obtained respectively. The relative values of these directions determine the shape of the element, and then the distance and direction of the center of the element (triangle) relative to the singular point are obtained. Use this data to construct matching variables:
Vi={Meta Basic Information, Meta Location Information}; I=0,1,2,3...
The basic primitive information A={Endpoint 1 type, with Endpoint 1 as the vertex angle, Endpoint 2 as the vertex angle, Endpoint 3 as the vertex angle}; Meta Location Information B={The distance from the center point of a meta to the singular point, and the direction of the center point of a meta to the singular point}. Note: In the basic primitive information, the order of the endpoints corresponds to their proximity to the center point, that is, the nearest endpoint is placed first and the farthest endpoint is placed on the surface.
From the matching variables, we can see that the basic information of a primitive is only related to the shape of the primitive itself, but not to its location and size. Meta-location information is related to the location of the meta-image and not to the shape of the meta-image. The reason for this is that in the case of limited fingerprint image area, two similar but different primitives (the exact same shape but different size in extreme cases) will have very different adjacent primitive shapes, which will be excluded when other primitives are matched.
This meta-structure matching algorithm can improve and improve the matching performance in at least three aspects. First, the matching of each feature point is transformed into the matching of primitives, which greatly reduces the amount of computation. Secondly, it can significantly reduce the rejection rate and misrecognition rate of fingerprint matching. Thirdly, as long as the singularity (only related to fingerprint features) is calculated accurately, it is robust to the distortion, translation and rotation of fingerprint images.

epilogue
Embedded automatic fingerprint recognition system uses the more advanced fingerprint automatic recognition algorithm at present, and has made some significant improvements in automatic matching technology. Especially, using the idea based on primitive matching, the rejection rate and misrecognition rate of fingerprint recognition are very low, the matching speed is very fast, and it is especially suitable for the application of large-capacity fingerprint library. After the system design is completed, the actual application tests show that the automatic fingerprint recognition of the system is fast, accurate and completely practical. ^

Reference
1 Li Qunfang. Single-chip computer and interface technology. Central China University of Technology Press, October 1998: 10~53
2 Yin Yilong, Zhan Xiao4, Tan Taizhe, Ning Xinbao. Fingerprint enhancement algorithm based on Gabor function and its application. 2003,14(3): 484~489
3 Jane Bing, Zhuang Zhenquan, Li Haiying, Wang Ruibin. Fingerprint detail extraction algorithm based on ridge line tracking. Journal of Circuits and Systems, September 2001 Volume 6, Issue 3