Face Alignment Using Active Shape Model And Support Vector Machine
This work addresses face alignment accuracy for applications like facial feature localization, but it is incremental as it builds on existing ASM methods.
The paper tackled the problem of low accuracy in classical Active Shape Models for face alignment by proposing four improvements, including combining Sobel filters with 2-D profiles, applying Canny edge enhancement, using SVM for landmark classification, and automatically adjusting profiles based on image size, resulting in far better performance on Caltech and imm_face databases.
The Active Shape Model (ASM) is one of the most popular local texture models for face alignment. It applies in many fields such as locating facial features in the image, face synthesis, etc. However, the experimental results show that the accuracy of the classical ASM for some applications is not high. This paper suggests some improvements on the classical ASM to increase the performance of the model in the application: face alignment. Four of our major improvements include: i) building a model combining Sobel filter and the 2-D profile in searching face in image; ii) applying Canny algorithm for the enhancement edge on image; iii) Support Vector Machine (SVM) is used to classify landmarks on face, in order to determine exactly location of these landmarks support for ASM; iv)automatically adjust 2-D profile in the multi-level model based on the size of the input image. The experimental results on Caltech face database and Technical University of Denmark database (imm_face) show that our proposed improvement leads to far better performance.