Abstract: This paper proposes a fingerprint recognition system based on DSP and Ethernet. The system design scheme is presented, and the process of receiving and processing the acquired fingerprint image data via the DSP chip is detailed. The data is then transmitted to the local area network through the interface with the Ethernet control chip RTL8019, thus achieving fingerprint recognition over a wide area. Keywords: Fingerprint recognition, Ethernet interface 1 Introduction Human fingerprints, with their uniqueness and immutability, have become one of the most reliable and convenient methods for identity verification. With the development of computer and image technology, the application fields of fingerprint recognition are becoming increasingly widespread. Currently, fingerprint recognition technology has been successfully applied in criminal investigation, household registration management, and bank savings systems. The rapid advancement of network technology has turned the world into a global village, and how to effectively connect fingerprint recognition with the Internet, enabling people to quickly and accurately verify their identity using fingerprints in different locations, is also very important. This paper describes the application of a DSP system interconnected with a computer network to process and transmit fingerprint images. Since most domestic local area networks (LANs) use Ethernet, this paper mainly introduces the application of TI's fifth-generation product, the TMS320VC5402, and the RTL8019 Ethernet controller to achieve fingerprint image processing and transmission. 2. Basic Components of a Fingerprint Recognition System The following block diagram illustrates the basic components of a fingerprint recognition system. There are various methods for fingerprint acquisition, including using geometric optics theory followed by photoelectric conversion to directly convert the fingerprint into an electrical signal. Fingerprint image acquisition input involves converting the electrical signal into a digital signal. Specifically, the fingerprint indentation is converted into a digital image that the computer can recognize. A typical method is to digitize a 1×1 fingerprint image into a 512×512 pixel array (each pixel contains 8 bits). This digital image is then sent to an image processing system with a DSP chip at its core. Ethernet input: After processing, the fingerprint image is transmitted to the LAN via an Ethernet interface chip for matching and recognition against the fingerprint database of the recognition center, thus achieving remote fingerprint recognition. The entire system uses a DSP chip as its core for fingerprint image preprocessing and fingerprint feature extraction, and transmits the pre-identified fingerprint image to the local area network via an Ethernet interface chip. The following section focuses on the system design, fingerprint image processing method, and recognition method. 2.1 System Design The design of this system mainly involves the design of two interfaces: the interface method between "fingerprint image acquisition" and "fingerprint image processing," and the interface method between "fingerprint image processing" and "Ethernet transmission." 2.1.1 Interface 1 The DSP chip used in this system is a fifth-generation product manufactured by TI, the TMS320VC5402, capable of 16-bit fixed-point arithmetic. It has a high processing speed, 16K words of on-chip RAM and 4K words of on-chip ROM, and can be externally expanded. It also features two automatic buffered serial ports (BSP) and an HPI port for communication with external processors. It offers a high performance-to-price ratio, making it suitable for image processing. Figure 2 shows the specific connection method between the C5402 and image acquisition. (1) Memory Mapping Selection Mode: The CPU can typically operate in different modes depending on user needs, i.e., whether the program memory space is defined on-chip or off-chip, mainly determined by MP/MC. When MP/MC=1, the program memory space from 4000H to FFFFH is entirely defined as off-chip memory. When MP/MC=0, the program memory space from 4000H to EFFFH is entirely defined as off-chip memory, and the program memory space from FF00H to FFFFH is defined as on-chip memory. (2) Data Bus: The data bus I/O ports I/015-I/00 for image acquisition are connected to D15-D0 of C5402. At the same time, D15-D0 of C5402 is connected to I/015-I/00 of program memory and data memory. The OE and WE (read/write enable) of program memory and data memory are selected by MSTRB and R/W of C5402 after decoding by 74LVC139. The CE of program memory is selected by PS of C5402. The CE of the data storage is selected by decoding A15 and DS using a 74LVC139. (3) The address bus A15-A0 of the image acquisition is connected to the 16-bit address A15-A0 of the C5402. At the same time, A13-A0 of the C5402 is connected to A13-A0 of the data storage, while A16-A14 of the data storage is used as the control page selection. Therefore, the data storage is divided into 8 pages, each page is 16K words, and is mapped to memory space 0X4000 to 0X7FFF. A14-A0 of the C5402 is connected to A14-10 of the program storage, and A16 and A15 of the program storage are used as the control page selection. Therefore, the program storage is divided into 4 pages, each page is 32K words, and is mapped to memory space 0X8000-0XFFFF. 2.1.2 Interface 2 The RTL8019 is an Ethernet controller manufactured by Realtek Corporation of Taiwan. It adopts a 100-pin PQFP package. Its pins can be divided into power and clock pins, network media interface pins, bootstrap ROM and initialization EPROM interface pins, main processing interface pins, output indication and working mode configuration pins. It supports IEEE802.3 10Base, 10Base2, 10BaseT and occupies a considerable proportion in ISA bus cards; it supports 8-bit or 16-bit data bus and 16 I/O base address selection; it has a built-in 16K-word SARM for transmit and receive buffers; it has three selectable interface modes with the host, namely jumper mode, PnP mode and RT mode. In order to simplify the software and hardware design of DSP network, the jumper mode is selected in this paper. This mode is compatible with the early network controllers. Its port base, interrupt port, etc. are determined by switches or jumpers. (1) Mode selection The mode selection is determined by the JP and PnP pins. This system selects the jumper mode, and the JP pin is connected to a high level. To shield remote bootstrapping, the SMEMRB pin is connected to a high level. (2) Address bus: The RTL8019 has a total of 20 address lines. Here, 16 lines are used to address the I/O ports. That is, the A15-A0 of the DSP are connected to the SA15-SA0 of the RTL8019, and the rest are grounded. To make the I/O command valid, the address enable pin AEN is grounded. (3) Data bus: Select the 16-bit data bus, which requires the IOCS16 pin of the RTL8019 to be connected high. Since the interface level of the RTL8019 is 5V, while the bus level of the C5402 is 3.3V, a level conversion is required. (4) Interrupt and read/write control: The RTL8019 has a total of 8 interrupt outputs. Just select one of them as the interrupt request. The read/write status of the RTL8019 can be controlled by the I/O port control signals IS, IOSTRB, and R/W signals of the C5402 through 74LVCl39 decoding. To avoid the mismatch between the read and write speeds of RTL8019 and other chips, RTL8019 has a dedicated pin IOCHRDY for inserting wait cycles for host read and write commands. Therefore, connecting this pin to the READY pin of C5402 can prevent the read and write speed of C5402 from being too fast and incompatible with RTL8019. (5) Initialization Configuration: Use an extended output port of C5402 instead of a jumper to specify the I/O port base address, interrupt output port, and media type of RTL8019AS, and use an output signal as the reset signal of RTL8019AS. When RTL8019AS finishes resetting, it samples these configuration pins and initializes its internal configuration register according to the pin status. After the RTL8019 reset initialization, some contents of the register must also be initialized before data can be received and sent. Its built-in 16K word SRAM can be used for receiving and sending buffers. The buffer can be divided into 64 pages, with page numbers ranging from 0X40 to 0X80, and each page is 256 bits. The page range of the receive buffer is set by the PSTART and PSTOP registers, and the page range of the transmit buffer is set by the RSAR0,1 and RBCR0,1 registers. 2.2 Image preprocessing process The purpose of image preprocessing is to improve the quality of the input image and convert it into a clear dot and line image. The extraction and recognition of image features require high accuracy, so the image preprocessing process is relatively complex, specifically including fingerprint image enhancement, ridge enhancement and restoration, and binarization. Image enhancement includes smoothing, sharpening and filtering processes. This technique was perfected by O'Gorman and Nickerson in 1989 and will not be described in detail here. The following focuses on the enhancement and restoration of ridges. This process mainly includes two parts. The first part is to calculate the direction of the image from the input fingerprint image. The second part is to filter, enhance and restore the image using the direction calculated above. (1) Calculation of image direction As shown in Figure 4, the direction of a point (i,j) in the image is denoted as D(i,j). Here, Sd is calculated first, which is the sum of different gray values ​​along the direction d. In the above formula, f(i,j) and fd(ik,jk) represent the corresponding gray values ​​of pixels (i,j) and (ik,jk), respectively. (ik,jk) represents the k-th pixel along direction d from (i,j), n represents the number of pixels calculated, and N represents the number of directions used. We take N=16 directions and n=8, meaning 8 neighboring points are taken in each direction. The direction D(i,j) of point (i,j) takes the minimum value of Sd. From the above formula, we can conclude that the overall change in gray value is smallest along the ridge direction and largest perpendicular to the ridge direction. Therefore, the direction D(i,j) of pixel (i,j) is the direction in which the gray value is always at its maximum in the image. Furthermore, when calculating the gray value of points in each direction, the position of the point in each direction is obtained by rotating the coordinates of the point in the original 0 direction. If we calculate the coordinates of the i-th point in the 1 direction, then α represents the angle between the 0 and 1 directions. The direction of a region can be calculated by the direction of each pixel in a region. The direction of the region is the same as the direction of the pixel with the highest probability in the region. (2) Image processing method: Observation shows that the directions between the interconnected regions in the whole fingerprint image do not change drastically. Therefore, it can be judged that any region with a drastic change is a noise region. Then the direction of the region is replaced by the direction of the region with the highest probability in the surrounding regions. Generally, this operation is repeated twice, and about 70-80% of the noise can be removed. Moreover, the image enhancement and restoration effect is good. However, this method should not be used more than twice when the direction calculation of the region is not very accurate. Otherwise, some image distortion will occur. Binarization is to convert the image into a binary image with values ​​of 0 and 1. The average value of each block of the image is used as a threshold. According to this threshold, if the pixel value is greater than the threshold, it is 1. If the pixel value is less than the threshold, it is 0. Finally, fingerprint features are extracted. 2.4 Recognition method: After the processed image is transmitted to the local area network through the Ethernet interface chip, it is matched and recognized with the fingerprint pattern in the fingerprint database of the recognition center. Fingerprint patterns in the fingerprint database are classified according to detailed features. This reduces the number of comparisons between the pre-identified fingerprint and the fingerprint database, greatly saving recognition time and improving recognition accuracy. The following formula gives the comparison similarity matching function: S = (P**2/(M*N)) (4) Where S represents the matching degree, P represents the number of detail pairs of the fingerprint image, M represents the number of details of the pre-identified fingerprint image, and N represents the number of details of the fingerprint images in the fingerprint database. If the details of the pre-identified fingerprint image are similar to the details of the fingerprint images in the fingerprint database (within the allowable range), they are called a pair. Thus, the one with the highest matching number can be used as the candidate fingerprint. 3 Conclusion The fingerprint recognition system composed of DSP and Ethernet control chip can meet various requirements in terms of fingerprint image processing speed and quality. In particular, with an Ethernet interface, it can form a high-speed local area network with a PC via twisted pair or coaxial cable, expanding the application scope of the fingerprint recognition system. If the PC is connected to the Internet and different external hardware and software are set up, the application areas of fingerprint recognition can be further extended.