Subpixel Line Localization With Normalized Sums of Gradients and Location Linking With Straightness and Omni-Directionality

This paper presents a method to localize line locations with subpixel accuracy and a method to link the locations based on a linking distance. This paper first proposes a subpixel line localization method based on normalized sums of gradients (NSG) calculated by dividing pixel sum of gradients by the sum of gradient lengths within the pixel neighborhood. The proposed NSG method is compared with current state-of-the art based on a Taylor series approximation of intensity surface and the normal vector derived from the Hessian matrix. Comparative experiments for subpixel line localization methods were performed with simulated and natural images and confirmed the proposed subpixel localization method provided superior accuracy and faster localization under most combinations of varying line width and noise strengths than the state-of-the art localization method. The proposed linking method was also designed to have more straightness and omni-directionality than a current state-of-the art method. Experimental comparison of linking methods confirmed the proposed linking method provided superior linking quality than current state-of-the art.