Abstract
Conference Title: 2016 IEEE International Conference on Electro Information Technology (EIT) Conference Start Date: 2016, May 19 Conference End Date: 2016, May 21 Conference Location: Grand Forks, ND, USA Acne is a common chronic skin disease involving blockage and/or inflammation of hair follicles and their accompanying sebaceous gland. Acne can present as non-inflammatory lesions, inflammatory lesions, or a mixture of both, affecting mostly the face but also the back and chest. Detecting the different types of acne lesions is important in both diagnosis and management. According to acne face mapping, presence of acne in various parts of the face or body has different indications for disease. In this paper, we present several image segmentation methods to detect acne lesions and machine learning methods used to distinguish different acne lesions from each other. Our results illustrated that among texture analysis, k-means clustering, HSV model segmentation techniques, two level k-means clustering outperformed the others with an accuracy of about 70%. In addition, the accuracy of differentiating acne scarring from active inflammatory lesions is 80% and 66.6% for fuzzy-c-means and support vector machine method, respectively. Finally, the performance accuracy of classifying normal skins from detected acnes is 100% using fuzzy-c-means clustering.