Fast Implementation Of Morphological Operations Using Binary Image Block Decomposition

Morphological transformations are commonly used to perform a variety of image processing tasks. However, morphological operations are time-consuming procedures since they involve ordering and min/max computation of numbers resulting from image interaction with structuring elements. This paper presents a new method that can be used to speed up basic morphological operations for binary images. To achieve this, the binary images are first decomposed in a set of non-overlapping rectangular blocks of foreground pixels that have predefined maximum dimensions. Then off-line dilation and erosion of all rectangular blocks are arbitrary obtained and stored into suitable look-up array tables. By using the look up tables, the results of the morphological operations to the rectangular blocks are directly obtained. Thus, first all image blocks are replaced by their look-up array tables. Then the morphological operations are applied only to the limited number of the remaining pixels. Experimental results reveal that starting from a block represented binary image morphological operations can be executed with different types of structuring elements in significantly less CPU time. Using the block representation, we are able to perform dilation 16 times faster than non-fast implementations and 10 times faster than an alternative fast implementation based on contour processing. Significant acceleration is also recorded when using this approach for repeated application of dilation (for 10 iterations, dilation using the block representation is over 20 times faster than non-fast implementations and over four times faster than using the fast contour based approach).