Software for Automatic Evaluation of Stretch Marks

Considering that the current protocols in the evaluation of stretch marks are based on the visual clinical examination of the striated skin and in the measure of the same, this article aims to assist the professionals of the dermatology through data for decision making in the diagnosis and follow-up of the treatment of the striations. In order to do so, a software supported by computational vision and machine learning techniques was developed, aiming to collect images and analyze them to identify striae. From a database of images annotated by specialists in dermatology, color and texture attributes of the segments were extracted, which were later classified by the Random Forest algorithm, which presented a 99.92% correct classification of the striae. In addition, metrics were analyzed that indicated that although the streak superpixels were correctly classified, other classes were classified incorrectly.