Developing color-guided multi-seed region growing method for automated acne vulgaris detection and classification

Detection and classification of Acne vulgaris remain challenging due to the complex characteristics and severity of acne, as well as the variations in human skin types. Dermatologists typically detect acne manually using grading techniques, which are inefficient as they are time-consuming and subjective. Therefore, it is necessary to develop methods for detecting and classifying acne from skin images. In this article, acne identification and classification are developed through segmentation in the YCbCr color space using the region growing method. Seed points for the region growing method are provided in the form of multi-seed points, which are selected from a database built from 23 training images featuring three types of acne with various skin backgrounds. The multi-seed points consist of a set of Cb and Cr color components from the skin, papule acne, pustule acne, and nodular acne. Each data group within the multi-seed points serves as the precursor for forming regions corresponding to the segmented area types. The multi-seed points are chosen to ensure balance across all types of regions to be segmented and are distributed throughout the evaluated images. The method was tested on 13 cheek skin images of acne sufferers. The region growing method was provided with an initial dataset, consisting of 2 data points for each of the 4 different skin types selected from the multi-seed points. Out of 54 acne lesions diagnosed by dermatologists, 89 %, or 48, were successfully segmented as acne. Furthermore, acne classification using KNN showed that of the 48 segmented acne lesions, 43, or 90 %, were correctly classified. Multi-seed points as the starting points for segmentation using the region growing method demonstrated high accuracy