Make biometric-based person identification system applicable

• Main Author: Chan, Yuen Loong.
• Other Authors: Li Fang
• Format: Final Year Project
• Language: English
• Published: 2010
• Subjects: DRNTU::Engineering
• Online Access: http://hdl.handle.net/10356/39755
• Institution: Nanyang Technological University

This report has explored and developed ways of reducing the time and resource (System Memory) consumption of the palm-print identification program. With the possibility of implementing the palm-print program on a mobile system, there will be hardware limitations such as processing power and system memory of the mobile device. Coupled with the need for an improved user experience, there is a need to reduce the time and resource of the program. This report focused on the existing program’s implementation of loops and the usage of data structure to achieve its desired results. Through re-implementation of the code, the obtained results will be compared to the existing results. A final evaluation was done based on the results obtained to verify the improvements to the time and resource utilization. The existing program was found to contain numerous resource inefficiencies. The use of arrays for certain operations increased the system memory requirement while causing space and time wastage. They account for 900,000 Kilo-Bytes of memory during the identification process rendering computer systems with less than 1GB of Random Access Memory (RAM) to be unable to execute the code. In addition, the use of arrays was not rightfully justified as random access was not required. Therefore, the double linked-list structure was proposed since the code consists mainly of insert operations. Another significant cause to the time efficiency is the loop structure where redundant file Input/output (I/O) occurred. 2 new system flow design, Model Repository (MRP) and Low RAM (LowRam), had been proposed to replace the previous system flow design. The new system reduced complexity significantly by up to 98%. The new data structure performed up to 88% faster than the previous implementation during actual real-life scenario where model and input files are larger than 100 each.