Skin image processing and classification
Date of Issue2016
School of Electrical and Electronic Engineering
National Skin Centre
Melasma is a skin pigmentation disease that can lead to substantial embarrassment and distress in humans’ daily life. For assessment of melasma, current diagnosis is conducted by observing melasma pigmentary area and extent of pigmentation with conventional visual assessment methods, which is exceedingly inconsistent and subjective. Thus, a computerized scoring method is highly demanded to eliminate the biased assessment by producing a standard score. This report proposes an automated scoring method for melasma pigmentary area segmentation and classification utilizing reaction-diffusion based level set model together with local entropy thresholding method. In this level set model, a diffusion term is used to regularize the level set function while a reaction term with anticipated sign property is used to force zero level set towards desired locations. Then anticipated boundaries are filtered out by the local entropy thresholding method where boundaries with higher overall local entropy are extracted and misclassified regions are excluded. Eventually, the target object (melasma pigmentary area in our case) and the background (normal skin area) can be identified. Experimental results indicate that the proposed method can produce a reasonable outcome. This work provides a new approach for further investigations on melasma image segmentation using level set method.
Final Year Project (FYP)
Nanyang Technological University