We present a new method, which allows successful application of watershed transformation for images containing small objects. The method can be divided into two stages. During the first stage of segmentation a special gradient operator is applied to get the gradient of the image. It gives us a possibility to increase the resolution of the gradient doubly in comparison with the resolution of the original image. We can achieve this by approximation of the first-order derivative not only in image pixels, but also between pixels. During the second stage the author's modification of watershed transformation is applied. This modification enables us make optimal use of the image gradient obtained at the first stage. The method to be introduced was probed on a large amount of images with small details, and showed better results than the traditional watershed algorithm.
|Number of pages||4|
|Journal||Proceedings - International Conference on Pattern Recognition|
|Publication status||Published - 2004|
|Event||Proceedings of the 17th International Conference on Pattern Recognition, ICPR 2004 - Cambridge, United Kingdom|
Duration: 23 Aug 2004 → 26 Aug 2004