Simple mathematical operations based classification of the light color values of the images for skin cell detection.

Abstract

Skin disorders are identified visually by expert medical doctors before the detailed examination stage. In addition, some automatic computer-aided diagnostic methods have been developed to help the doctor at the visual examination stage. These methods are mostly based on skin color classification because the color change in skin is important for medical doctors. However, proposed classification methods are based on complex mathematical equations. Therefore, methods proposed in the literature are difficult to understand by medical doctors and these methods cannot be used without a computer. In this study, a new classification method that based on simple mathematical operations and can be used without a computer is proposed. In the first stage of the study, relation between the primary light color values of the images was inspected visually and visibly obvious differences were determined to create new attributes. Then the classification was made in the following way If primary light color order is R>G>B and then the R/G ratio is between the determined range, the image sample was classified as skin image. Determined ratio (R/G ratio range of skin samples) was determined by the method of decision trees. Finally, the obtained new attributes were classified by using well-known classifiers again and the classification accuracy (99.2%) of the proposed method was approved.