Paper Contents
Abstract
This paper introduces a novel approach to facial expression analysis by identifying both common and specific patches crucial for discriminating among various expressions. Leveraging the insight that only select facial regions, such as around the mouth or eyes, exhibit activity during expression disclosure, the study endeavors to uncover these discriminative patches. A two-stage framework, known as multitask sparse learning (MTSL), is proposed to efficiently locate these patches. The initial phase of MTSL involves amalgamating expression recognition tasks to pinpoint common patches essential across expressions. Each task focuses on identifying dominant patches corresponding to specific expressions. In the subsequent stage, two interrelated tasksfacial expression recognition and face verificationare integrated to learn specific facial patches unique to individual expressions. These two-stage learning processes operate on patches sampled via a multiscale strategy. Moreover, this paper emphasizes the critical role of accurate facial landmark detection in identifying salient facial patches. The proposed framework relies on the extraction of discriminative features from these identified patches, contributing significantly to robust expression recognition. Furthermore, an automated learning-free method for facial landmark detection is introduced, optimizing execution time while maintaining performance akin to state-of-the-art landmark detection techniques. The system's efficacy in low-resolution images is highlighted, showcasing consistent performance across various resolutions. Experiments conducted on renowned facial expression databases, including CK+ and JAFFE, demonstrate the effectiveness of the proposed methodologies. Through these experiments, the framework showcases superior performance in expression recognition, outperforming existing state-of-the-art methods. The integration of facial landmark detection and feature selection from salient facial patches contributes substantially to the system's accuracy and adaptability, particularly in varying resolutions. KeywordsFacial expression analysis, facial landmark detection, feature selection, salient facial patches, lowresolution image
Copyright
Copyright © 2024 Muskan. This is an open access article distributed under the Creative Commons Attribution License.
