Pixel-based skin segmentation in psoriasis images.

Area of Science:

• Medical Imaging
• Computer Vision
• Dermatology

Background:

• Psoriasis image analysis requires accurate skin segmentation.
• Existing segmentation methods need evaluation for psoriasis-specific applications.

Purpose of the Study:

• To compare various skin segmentation techniques for psoriasis images.
• To identify optimal color spaces and decision boundaries for improved skin detection.
• To evaluate the effectiveness of histogram-based Bayesian classifiers against explicit thresholding.

Main Methods:

• Investigated explicit skin thresholding with CbCr, HS, and rgHSV decision boundaries.
• Applied histogram-based Bayesian classifier to generate skin probability maps (SPMs) across five color spaces (YCbCr, HSV, RGB, XYZ, CIELab).
• Utilized decision trees and Otsu's thresholding to create binary skin maps (SMs) from SPMs, followed by morphological operations for refinement.

Main Results:

• The histogram-based Bayesian classifier significantly outperformed explicit thresholding for psoriasis images.
• The CbCr decision boundary demonstrated superior performance compared to HS and rgHSV.
• SPMs from Cb, Cr, H, and B-CIELab color bands produced the most effective binary skin maps.

Conclusions:

• The histogram-based Bayesian classifier is a highly effective method for skin segmentation in psoriasis images.
• Optimal performance is achieved using the CbCr decision boundary and specific color bands within CIELab and YCbCr spaces.
• This research provides a robust framework for automated psoriasis image analysis.