Research on hair removal algorithm of dermatoscopic images based on maximum variance fuzzy clustering and optimization Criminisi algorithm. (September 2022)

• Record Type: Journal Article
• Authors: Song, Xiaowei; Guo, Shuli; Han, Lina; Wang, Li; Yang, Wentao; Wang, Guowei; Anil Baris, Cekderi
• Is Part Of: Biomedical signal processing and control. Volume 78(2022)
• Publication Date: 2022-09
• Subjects: Melanoma -- Hair remove -- Maximum variance -- Fuzzy clustering -- Criminisi algorithm; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.103967
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Hair region can be extracted more accurately based on maximum variance fuzzy clustering algorithm. Perfect hair details are repaired based on improved Criminisi priorities, matching criteria and matching paths. Presents a qualitative and quantitative evaluation criterion of hair restoration algorithm. Abstract: Hair removal is one of the significant challenges before applying the automatic segmentation and classification process for skin lesions in a computer-aided diagnosis system for melanoma. Under this background, an efficient hair removal algorithm for dermatoscopic images, is offered in this paper. First, grayscale and wave valley detection were performed on the image, and the hair regions determined by maximum variance fuzzy clustering based on the color and shape characteristics of the image, were corrected by region growth, and were repaired by criminisi algorithm with improved priority and matching criteria. Second, the improved priority is combination of the traditional one and characteristics of structural strength and data coherence. The color weight is introduced into the sample block matching criterion, and the ant colony algorithm is used to optimize the search path of the matching block. Third, we also proposed a qualitative and quantitative method evaluating hair extraction-repair, and five typical hair repair experiments are shown in details with validated verification algorithm by using collected and ISIC 2019 data sets. The last, Experimental results show that the accuracy of our hair detection experiments in the ISIC2019 dataset, can be improved by 2–7% with the accuracy of hair repair experiments improved by 2–5% on average.