A Drawing Accuracy Evaluation Method with Localized Normalized Cross-Correlation for Portrait Drawing Learning Assistant System

Nowadays, portrait drawing has gained popularity from various advantages in human sentiment cultivations and digital technology progresses including a tablet or a PC. However, it is still difficult for novices to get started from scratch. Previously, we have developed the Portrait Drawing Learning Assistant System (PDLAS) to assist portrait drawing by providing auxiliary lines as a reference. They are generated by running OpenPose and OpenCV to extract facial features. However, a method of evaluating the drawing accuracy is not implemented in PDLAS. In this paper, we present a drawing accuracy evaluation method of calculating the Normalized Cross-Correlation (NCC) for each face part to quantify the similarity between a user draw and its original face image. By localizing the range for each part, the result is improved. In the evaluation section, we analyze the scores assigned to different facial features by the NCC and identify the issues revealed by the scores for each user. Furthermore, we provide suggestions for improvement based on the weaknesses found in the users. Thus, we can provide specific feedback to the users.