PDE-based unsupervised repair of hair-occluded information in dermoscopy images of melanoma

Automated Digital Hair Removal by Threshold Decomposition and Morphological Analysis

“…For image (g), our method creates less halos around removed hairs than both DullRazor and Huang et al For image (h), our method removes considerably more hair than both previous methods. Figure 3 compares our results with Xie et al [30] and Huang et al While our method removes more hairs than Xie et al, it also removes a small fraction of the skin texture (which is suboptimal). In contrast, Huang et al removes all hairs but also massively blurs out the skin pattern.…”

“…However, the found edges are thresholded, so hair close to skin color is hard to find. Top-hat filtering has also been used to find hairs [30]. Huang et al find hairs by multiscale matched filtering and hysteresis thresholding and remove these by region growing with PDE-based inpainting [13].…”

“…However, this method is quite slow (minutes for a typical dermoscopy image). VirtualShave finds hairs by top-hat filtering, similar to [30], and uses three density, sphericity, and convex-hull sphericity metrics to separate thin elongated structures (hairs) from other high-contrast details (false positives) [12]. Abbas et al propose to find hairs by a derivatives-of-Gaussian (DOG) filter [1,2].…”

“…(a) fitting lines in a finite number of orientations and checking the length of the longest such line [19]; (b) using principal component analysis to find if the major-to-minor eigenvalue ratio exceeds a given threshold [18]; and (c) computing an elongation metric comparing a component's skeleton-length with its area [30]. Xie et al argue that (a) and (b) are ineffective, as they favor mainly straight hairs and yield false-negatives for highly curled hairs [30]. To alleviate this, they use an elongation metric equal to the ratio of the area C i to the squared length of the 'central axis' of C i .…”

“…Several digital hair removal (DHR) methods address this by finding and replacing such hairs by plausible colors based on surrounding skin. Despite much work in this area [19,30,15,12,2,13], DHR methods are challenged by hairs which are thin, entangled, of similar contrast or color to underlying skin, or overlaid on a highly-textured skin structure.…”