DA-FER: Domain Adaptive Facial Expression Recognition

Facial expression recognition (FER) is an important field in computer vision with many practical applications. However, one of the challenges FER dealing small sample data, where number samples available for training machine learning algorithms limited. To address this issue, a domain adaptive strategy proposed paper. The approach uses public dataset sufficient as source and target domain. Furthermore, maximum mean discrepancy kernel embedding utilized to reduce disparity between data samples, thereby enhancing accuracy. Domain Adaptive Expression Recognition (DA-FER) method integrates SSPP module Slice fuse features different dimensions. Moreover, retains regions interest five senses accomplish more discriminative feature extraction improve transfer capability network. Experimental results indicate that can effectively enhance performance recognition. Specifically, when self-collected Selfie-Expression used domain, datasets RAF-DB Fer2013 are improved varying degrees, which demonstrates effectiveness method.