A Sign Bit Only Phase Normalization for Rotation and Scale Invariant Template Matching

Using Symmetric Phase Only Matched Filter (SPOMF) to locate the position of a known object in static pictures or real-time streaming pictures has been proved to be a reliable method. A key issue of SPOMF is the normalization of the phase of the complex matrix. In this paper, several different solutions for phase normalization are introduced and compared. Starting from the CORDIC solution, experimental results indicate that phase angle quantization shows little influence to the quality of the phase normalization but can greatly increase the performance. The method is to take some most significant bits of the real part and the imaginary part of the complex numbers to represent the phase angles. Based on the MatLab simulation results, a highly efficient solution that is called Sign Bit Only Phase Normalization is developed. This solution only uses the sign bit to represent the phase angles so that all the complex numbers can only have 4 possible phase angles. In that case, the Sign Bit Only solution while being much easier to implement still shows acceptable results, in which the peak magnitude in the correlated image is decreased to about 90%. A mathematical explanation that proves the reliability of this solution and experimental results are also given in this paper. Keywords— Image Registration; FFT; SPOMF; Phase Normalization