Digital hair removal by deep learning for skin lesion segmentation
作者：
Highlights：
• Proposed a deep learning model for accurate segmentation of SKIN lesions through removal of hair strands.
• Employed a U-Net trained on manually created hair mask dataset to segment hair from skin lesion images.
• Used a combination of Gated convolution and SN-PatchGAN to inpaint the hair gap.
• Proposed a non-referenced Structural Similarity Index (SSIM) method referred as the Intra SSIM to evaluate the hair removal and the inpainting process.
• Used ISIC 2018 dataset, to evaluate the performance of the proposed method and compared it with other state-of-the-art methods.
摘要
•Proposed a deep learning model for accurate segmentation of SKIN lesions through removal of hair strands.•Employed a U-Net trained on manually created hair mask dataset to segment hair from skin lesion images.•Used a combination of Gated convolution and SN-PatchGAN to inpaint the hair gap.•Proposed a non-referenced Structural Similarity Index (SSIM) method referred as the Intra SSIM to evaluate the hair removal and the inpainting process.•Used ISIC 2018 dataset, to evaluate the performance of the proposed method and compared it with other state-of-the-art methods.
论文关键词：Dermoscopy,Digital hair removal,Skin lesion segmentation,Deep learning
论文评审过程：Received 7 July 2020, Revised 2 April 2021, Accepted 18 April 2021, Available online 28 April 2021, Version of Record 10 May 2021.
论文官网地址：https://doi.org/10.1016/j.patcog.2021.107994
