Research On Improved Watershed Algorithm

To reduce the over-segmentation phenomenon caused by noise and texture in classical watershed algorithm,a new improved watershed algorithm is proposed.The new algorithm consists of two parts,including image optimization and results correction.In the stage of image optimization,firstly,image object details are abstracted,morphological structuring elements combining object structure details are designed,and a corresponding non-linear filter is used to smooth image,which can remove noise data while remain object edges information effectively.Then,the RGB vector gradient of filtered image is computed as initial gradient of watershed transform,which guarantees that the final waterlines are generated at the correct object edges.To further improve the contrast ratio between edges data and noise data,multi-scale reconstruction technology is adopted to operate the gradient image.In the end,markers are extracted in reconstructed gradient using extended H-minima technology to modify initial gradient,and watershed algorithm is applied,which produces more ideal segmentation results.In the phase of results correction,the adjacent region merging rule based on global maximum similarity is put forward,which makes final object regions more in line with the human vision.Simultaneously,the merging constraint condition is designed,which halves the complexity of algorithm.Using image dataset BSDS300 to train and analyze the algorithm proposed,and results show that the new algorithm proposed in this paper has high noise immunity,precise contours location,and is better than other segmentation algorithms.