Analysis of Facial Features UsingGaussian Steerable

A novel approach for the analysis of facial features is presented. Instead of analyzing facial features with global models (such as the popular deformable templates) we perform a local analysis of the facial features. This local analysis is integrated in a system for diagnosis support of patient aaicted with single{sided facial paresis. This kind of paresis produces asymmetries in the patient's face, specially in the eye and mouth regions. Moreover, the asymmetries are ampliied when special mimic exercises are performed. In this system we apply Gaussian averaging masks, which are used as a steerable lter, to extract information about facial features with high precision to detect asymmetries inside the face. The system is composed of three successive functions: facial feature localization, local orientation analysis, and facial asymmetry evaluation. In an initial step the eye and mouth regions are localized. Then Gaussian averaging masks are applied to characterize the orientation information in the neighborhood of the corners of the eyes and mouth. This orientation information is analyzed and used for the diagnosis supporting task. Real experimental results show that this technique is promising. Abstract. A novel approach for the analysis of facial features is presented. Instead of analyzing facial features with global models (such as the popular deformable templates) we perform a local analysis of the facial features. This local analysis is integrated in a system for diagnosis support of patient aaicted with single{sided facial paresis. This kind of paresis produces asymmetries in the patient's face, specially in the eye and mouth regions. Moreover, the asymmetries are ampliied when special mimic exercises are performed. In this system we apply Gaussian averaging masks, which are used as a steerable lter, to extract information about facial features with high precision to detect asymmetries inside the face. The system is composed of three successive functions: facial feature localization, local orientation analysis, and facial asymmetry evaluation. In an initial step the eye and mouth regions are localized. Then Gaussian averaging masks are applied to characterize the orientation information in the neighborhood of the corners of the eyes and mouth. This orientation information is analyzed and used for the diagnosis supporting task. Real experimental results show that this technique is promising.