Occlusion due to hair in dermoscopic images affects the diagnostic operation and the accuracy of its analysis of a skin lesion. Also, dermis hair has the following different characteristics: thin; overlapping; faded; of similar contrast or colour to the underlying skin; and obscuring/covering textured lesions. These make digital hair removal (DHR), which involves hair segmentation and hair gap inpainting, a challenging task. Thus, traditional hard-coded threshold-based hair removal methods are not effective, resulting in over-removal which loses important information of the skin lesion, or under-removal which cannot remove the hair effectively. In this paper, we propose a deep learning approach to DHR based on U-Net and a free-form image inpainting architecture. In hair segmentation, a well-labelled dataset is created and used to train U-Net in order to obtain accurate hair masks. In inpainting, a free-form image inpainting architecture (i.e., Gated convolution and SN-PatchGAN) which has been trained on millions of images is used to inpaint any hair gaps. We also propose an evaluation method to analyze the effect of hair removal based on a single dermoscopic image, named intra structural similarity (Intra-SSIM). The process of DHR is repeated until there is no change in the average value of Intra-SSIM. Using the ISIC 2018 dataset, the performance of the proposed method is shown to be better than other state-of-the-art methods.

皮肤镜图像中由于毛发引起的阻塞会影响诊断操作及其对皮肤病变的分析准确性。另外，真皮头发具有以下不同特征：稀疏；重叠; 褪色 与下层皮肤具有相似的对比度或颜色；并遮盖/覆盖有纹理的病变。这些使数字脱毛（DHR）成为一项艰巨的任务，其中涉及到头发的分割和头发间隙的修补。因此，传统的硬编码的基于阈值的脱毛方法是无效的，导致过度去除会丢失皮肤病变的重要信息，或者导致去除不足而无法有效去除头发。在本文中，我们提出了一种基于U-Net和自由格式图像修复架构的DHR深度学习方法。在头发分割中 创建一个标签良好的数据集，并将其用于训练U-Net，以获得准确的发膜。在修复中，已经对数百万个图像进行训练的自由形式的图像修复体系结构（即Gated卷积和SN-PatchGAN）用于修复任何毛发。我们还提出了一种评估方法，用于基于单个皮肤镜图像分析脱毛的效果，称为内部结构相似性（Intra-SSIM）。重复DHR的过程，直到Intra-SSIM的平均值没有变化为止。使用ISIC 2018数据集，该方法的性能显示出优于其他最新方法。经过数百万张图像训练的门控卷积和SN-PatchGAN用于修补任何毛发间隙。我们还提出了一种评估方法，用于基于单个皮肤镜图像分析脱毛的效果，称为内部结构相似性（Intra-SSIM）。重复DHR的过程，直到Intra-SSIM的平均值没有变化为止。使用ISIC 2018数据集，该方法的性能显示出优于其他最新方法。经过数百万张图像训练的门控卷积和SN-PatchGAN用于修补任何毛发间隙。我们还提出了一种评估方法，用于基于单个皮肤镜图像分析脱毛的效果，称为内部结构相似性（Intra-SSIM）。重复DHR的过程，直到Intra-SSIM的平均值没有变化为止。使用ISIC 2018数据集，该方法的性能显示出优于其他最新方法。