Vectorization of constant-time gray-scale morphological processing algorithm using AltiVec

Mathematical morphology is widely used in the area of signal and image analysis. It has enormous applications in detection of weak targets, object recognition and feature extraction, edge detection, image enhancement and many more. However, its use has long been hampered by its algorithmic complexity as the size of the structuring element or image grows. With the trend toward larger images and proportionally larger structuring element, the need for a fast and more efficient morphological processing algorithms become pressing. The performance of morphological processing algorithm can be speedup using AltiVec vetor processing unit. But the AltiVec programmers are frequently disappointed to discover that their AltiVec code is not much faster than their pre-existing scalar code. In some cases it may even be slower. In this correspondence, a new, simple yet much faster vectorized algorithm using AltiVec exhibiting constant-time complexity is described and analyzed. It is compared against the scalar version of the algorithm and other implementations. The proposed vectorized constant-time gray-scale morphological processing algorithm using AltiVec outperforms the 2-D naïve implementation, IXLib-AV library based implementation and the scalar version of constant-time morphological processing algorithm. It also exhibits the run time complexity independent of size of structuring element.