Drosophila image Segmentation using Marker Controlled Watershed

Automatic segmentation of biological image is an important step to understand a biological process. Moreover, the task is difficult due to ambiguous boundaries and intensity inhomogeneity. The Marker Controlled Watershed (MCW), was introduced for biological image segmentation as an alternative to the Watershed algorithm which suffers from oversegmentation. However, in the basic version of the MCW, the chosen markers directly impacts the segmentation performance. In this paper a new algorithm is proposed and applied for Drosophila image segmentation. The idea is to consider the characteristic of Drosophila image and use kernels of cells as objects markers. The Kernels are constructed using the Fiji software. The performance of the proposed algorithm is attested and compared to the basic version of the MCW. Results show the efficiency of the proposed algorithm for Drosophila image segmentation in term of the Dice coefficient and F-1 measure.