An Approach for Quantitative Evaluation of the Degree of Facial Paralysis Based on Salient Point Detection Junyu Dong 1 , Lijing Ma 1 , Qingqiang Li 1 , Shengke Wang 1 Li-an Liu 2 , Yang Lin 1 , Muwei Jian 1 1 College of Information Science and Engineering, Ocean University of China, Qingdao, Shandong, 266100, China 2 Qingdao Hiser Medical Center, Qingdao, Shandong, China dongjunyu@ouc.edu.cn, malijing66@163.com Abstract Facial paralysis can be on one or both sides of face and one side is much more common. This disease can impose significant psychological and functional impairment to patients. Traditionally, patients with facial paralysis are evaluated and examined by physicians based on manually measurement of certain difference between the two facial sides. In this paper, we describe a new approach for quantitatively estimating the degree of facial paralysis. We first determine key points based on salient point detection. Then the differences between the two facial sides are calculated. The proposed method is proved to be useful and effective in practice. 1. Introduction Facial paralysis is seriously deteriorating patients’ normal life. Meanwhile, traditional methods that solely depend on the physician’s diagnoses are time consuming and subjective. An accurate method for assessing facial nerve system is useful and necessary. There are more than twenty methods regarding facial paralysise assessment in the literature. Latest and frequently used methods include Nottingham system [1] , Toronto facial grading system (TFGS) [2-3] , facial nerve function index (FNFI), 1inear measurement index (LMI) [4] and House-Brackmann(H-B) [5] system. These methods have defaults in integration, feasibility, accuracy and reliability, and in general are not commonly employed in practice. In this paper, we describe a method which is used to evaluate the degree of one side facial paralysis caused by function disorder problems of facial nerves, and make a quantitative assessment of patients’ paralysis and health status in order to help the physicians to choose the most appropriate treatment scheme. The steps of our facial paralysis detection method is composed of face detection, salient points detection [6] , edge detection, K-MEANS clustering [7] , key points detection and the final evaluation. In our practice, it is proved to be feasible and time-saving. The main idea is as follows. We first find the salient points in the face region. Since the salient points include some points that cannot describe facial features, edge detection is used to discard these points. Then K-MEANS clustering is applied to classify the salient points into six categories which represent two eyebrows, two eyes, nose and mouth. Following this fourteen key points are found in the six facial regions respectively. They mostly represent the state of the disease. Finally, certain vertical distances are calculated to assess the degree of patient’s paralysis. 2. Pre-processing to facial images The facial images are captured by ourselves in the hospital and they are original, and some noises will be taken in to deteriorate the quality of the images. The noises are unavoidable and destructive. Thus, Pre- processing is necessary to remove the noises in order to simplify the subsequent processes. First, the images are change into gray ones, and then we apply certain filters to the images to remove the noises. 3. Estimating degrees of facial paralysis In order to estimate the degrees of the facial paralysis, the following steps are taken: (1) Face region detection: to restrain facial features in a small rectangle in images; (2) Salient points detection: to find all salient points of the facial features; (3) Edge detection: Susan algorithm is used to find the edges of facial features including two eyebrows, two eyes, nose and mouth; (4) K-MEANS clustering: to classify the salient points into six regions; (5) Find the fourteen key points; International Symposium on Intelligent Information Technology Application Workshops 978-0-7695-3505-0/08 $25.00 © 2008 IEEE DOI 10.1109/IITA.Workshops.2008.93 483