Variable-Weighted Grayscale Centroiding and Accuracy Evaluating

Abstract

Grayscale-weighted centroiding weights the pixel intensity to estimate the center of a circular image region such as a photogrammetric target or a star image. This is easy to realize with good robustness; however, locating accuracy varies with weight. There is no conclusion as to the relationship between weight and centroiding accuracy. To find this relationship and thus acquire the highest achievable accuracy of centroiding, one must first derive the general model of centroiding as follows. ( 1 ) Set weighting as a function that includes the grayscale intensity of each pixel and two undetermined parameters α and β. The centers were calculated from the new weights. ( 2 ) Design an accuracy evaluation method for centroiding based on a virtual straight-line target to obtain the accuracy of centroiding with various parameters. With these parameter values and accuracy records, the accuracy-parameter curve can be fitted. In the end, simulations and experiments are conducted to prove the accuracy improvement of this variable-weighted grayscale centroiding compared with traditional centroiding. The experimental result shows that the parameter α contributes significantly to accuracy improvement. When α = 1.5, locating accuracy is 1/170 pixels. Compared to centroiding and squared centroiding, the new locating method has a 14.3% and 10.6% accuracy improvement indicating that it improves locating accuracy effectively.