Fast Binary Dilation/Erosion Algorithm Using Kernel Subdivision

Abstract

Numerous algorithms have been proposed in the literature to speed up dilation/erosion operations. The motivation has been to reduce computational complexity by exploiting the structuring element and the image object properties. This paper presents a new algorithm for binary morphological dilation and erosion called the Kernel Sub-Division algorithm and discusses its implementation in the two dimensional case. It decomposes the n-dimensional structuring element, into several subsets and operates on the object contours in the image. The image characteristics are exploited by subdividing the object contours into bins while performing contour processing. The elegance of the algorithm lies in its retaining the correspondence to the output of the classical implementation with massive speed gain. The results of the algorithm on a statistically significant test set of images, showed that it performed five times better than the classical implementation for a 3x3 kernel. It also demonstrated a marginal rise in execution time with increasing size of the kernel.