| Summary: | Skin pigmentation disorders are widely spread around the world and can bring huge negative emotional impact on the patients. To meet the high demand of proper diagnoses and treatments of the non-tumorous skin pigmentation disorders, image processing and machine learning techniques are investigated in the thesis. In particular, two problems during the treatment process, namely the classification and the severity assessment of the pigmentation disorders, are addressed. In the classification problem, five most commonly seen skin pigmentation disorders are studied. A voting based probabilistic linear discriminant analysis algorithm (V-PLDA) is proposed to tackle the large within-class variance problem in the image dataset. In the severity assessment problem, a complete severity assessment system for a common pigmentation disease, melasma, is developed. An optimal hybrid thresholding algorithm (OHYTA) is also proposed to segment the melasma pigmentation region properly in the image. To the best of the author’s knowledge, this is the first study that applies computerized methods to the classification and severity assessment problems for the non-tumorous skin pigmentation image dataset. The proposed algorithms are tested on real-world image data collected by the National Skin Centre of Singapore and the experimental results are verified by the dermatologists. It is shown that the results obtained are satisfactory and the proposed classification and segmentation algorithms outperform other state-of-the-art methods.
|
|---|