Watershed Segmentation Based on Multiscale Morphological Fusion

Image segmentation based on watershed method always results in over-segmentation. To alleviate this problem, watershed segmentation based on multiscale morphological fusion is presented. The target image is filtered with multiscale structure element and many result images are obtained with the parallel processing. This paper does image fusion using wavelet transform and choose the fusion principles respectively for high frequency and low frequency coefficients. This algorithm can eliminate the noise while preserving the main contour of target. Then this paper adopts image maximum entropy to determine the initial threshold of watershed. Lastly, an effective region merging algorithm is proposed to improve segmentation result. Experiments show that the proposed algorithm is efficient