![]() | Only 14 pages are availabe for public view |
Abstract Since the U-Net architecture demonstrates outstanding performance in the field of medical image segmentation, it is considered the basis of many proposed works in skin lesion segmentation. Despite the popularity of U-Net, the existence of a single encoder in the U-Net structure whose function is to jointly extract local features and global context information does not handle the large variations of skin lesion. In addition, borderline pixels are hard to classify because it contains some pixels that belong to the lesion and other pixels that belong to the background region. Therefore, accurately classifying the pixels of this borderline is a difficult task because it contains pixels of both types. |