A Feature-Preserving Hair Removal Algorithm for Dermoscopy Images
December 2011
in “
Skin Research and Technology
”
hair removal dermoscopy images melanoma diagnosis matched filtering first derivative of Gaussian morphological edge-based techniques fast marching inpainting diagnostic accuracy texture-quality measure lesion texture computer-aided detection CAD systems Gaussian filter edge-based techniques inpainting
TLDR The algorithm effectively removes hair from skin images, improving melanoma diagnosis accuracy.
The document details an algorithm created to detect and remove hair from dermoscopy images, which is essential for accurate melanoma diagnosis. The algorithm involves a three-step process: initial hair segmentation using matched filtering with the first derivative of Gaussian (MF-FDOG), refinement with morphological edge-based techniques, and repair using a fast marching inpainting method. Tested on 100 dermoscopy images, the algorithm was evaluated against other techniques and demonstrated superior performance, achieving a diagnostic accuracy (DA) of 93.3% and a texture-quality measure (TQM) of 90%. These results suggest that the algorithm is highly effective at preserving lesion texture while removing hair, making it a valuable tool for computer-aided detection (CAD) systems in melanoma diagnosis. The study was supported by the National Textile University Faisalabad-37610 and the Higher Education Commission (HEC) of Pakistan.
