The watershed transform: definitions, algorithms and parallelization strategies
Fundamenta Informaticae - Special issue on mathematical morphology
The Image Foresting Transform: Theory, Algorithms, and Applications
IEEE Transactions on Pattern Analysis and Machine Intelligence
| Hi-index | 0.00 |
The watershed transform is the popular method of choice for image segmentation of Region of Interest (ROI) in the field of mathematical morphology. However, like other segmentation methods, it has important drawbacks that include sensitivity to noise (poor detection of low signal to noise ratio structures) and over (under)- segmentation if an optimal threshold is not found. In addition, most of the times, the Watershed Transform is applied on the gradient estimation of the raw image which gives typical watershed like topography. Most watershed based segmentation is done with the use of gradient estimation, and very little work has been done addressing problems caused by such estimation. The use of gradient estimation worsens the resolution at the output and also adds noise causing over-segmentation. Recently, a graph based approach called Image Foresting Transform (IFT) was developed to address image partition problems from seed pixels into a shortest path forest problem in a graph, whose solution can be obtained in linear time. The watershed transform is best implemented using the IFT algorithm with priority queue data structure. IFT watershed algorithm gives us an option to introduce the different Monotonically Incremental (MI) Application Specific Lexicographic Path Cost Formulation (ASLPCF) in its wave front propagation; otherwise no such introduction was possible with the Classical watershed transform. The premise of this paper is to implement for the first time a variant of the watershed transform as a shortest path forest problem by introducing modified expression of ASLPCF in its wavefront propagation to overcome the low resolution and over-segmentation problems. This is achieved by enabling the use of the watershed transform directly on the raw image, achieving in this way a higher resolution. A comparison is performed between the results obtained from the proposed algorithm and the classical watershed transform, which demonstrates the accuracy of the algorithm in image segmentation.