This thesis implemented a field programmable gate array (FPGA)-based face detector using a neural network (NN), as well as a bit-width reduced floating-point unit (FPU). An NN was used to easily separate face data and non-face data in the face detector. The NN performs time consuming repetitive calculation. This time consuming problem was solved by a Field Programmable Gate Array (FPGA) device and a bit-width reduced FPU in this thesis. A floating-point bit-width reduction provided a significant saving of hardware resources, such as area and power.<p>The analytical error model, using the maximum relative representation error (MRRE) and the average relative representation error (ARRE), was developed to obtain the maximum and average output errors for the bit-width reduced FPUs. After the development of the analytical error model, the bit-width reduced FPUs and an NN were designed using MATLAB and VHDL. Finally, the analytical (MATLAB) results, along with the experimental (VHDL) results, were compared. The analytical results and the experimental results showed conformity of shape. It was also found that while maintaining 94.1% detection accuracy, a reduction in bit-width from 32 bits to 16 bits reduced the size of memory and arithmetic units by 50%, and the total power consumption by 14.7%.

This thesis implemented a field programmable gate array (FPGA)-based face detector using a neural network (NN), as well as a bit-width reduced floating-point unit (FPU). An NN was used to easily separate face data and non-face data in the face detector. The NN performs time consuming repetitive calculation. This time consuming problem was solved by a Field Programmable Gate Array (FPGA) device and a bit-width reduced FPU in this thesis. A floating-point bit-width reduction provided a significant saving of hardware resources, such as area and power.<p>The analytical error model, using the maximum relative representation error (MRRE) and the average relative representation error (ARRE), was developed to obtain the maximum and average output errors for the bit-width reduced FPUs. After the development of the analytical error model, the bit-width reduced FPUs and an NN were designed using MATLAB and VHDL. Finally, the analytical (MATLAB) results, along with the experimental (VHDL) results, were compared. The analytical results and the experimental results showed conformity of shape. It was also found that while maintaining 94.1% detection accuracy, a reduction in bit-width from 32 bits to 16 bits reduced the size of memory and arithmetic units by 50%, and the total power consumption by 14.7%.