Locating cross-shaped image feature with subpixel accuracy based on NCCO

Normalized cross correlation operator (NCCO) was used for image matching to localize popular cross-shaped features in machine vision. Distribution feature of similarity function around peak zone was found to be four symmetry hyperboloid planes (FSHP). An approach of subpixel localization based on hyperboloid fitting algorithm (HFA) was presented according to the character of FSHP. A coarse pattern position was localized by one-dimension searching in x and y direction respectively to reduce the computational complexity. The discrete NCCO points around the coarse position were fitted to four hyperboloid planes in four quadrants respectively through least square fitting method. The point of four hyperboloid intersection was adopted as subpixel localization position. The HFA was realized on a microscopy vision system of MEMS device micro-assembly work-cell. Experimental results show that the subpixel localization precision of HFA can reach 0.25 pixel. It has greatly decreased computative cost under the comparisons with conicoid fitting algorithm and Gaussian fitting algorithm.