Cross-Cultural Expression Spotting Method Based on Optical Flow and Sliding Window

One of the anticipated attributes of artificial intelligence is its ability to possess human-like characteristics, particularly in analyzing and understanding facial emotions. The challenge of the video macro/micro expression Cross-Cultural Spotting (CCS) task lies in the subtlety, brevity, and difficulty of accurately pinpointing micro-expressions. This paper proposes a macro/micro expression spotting framework based on Main Directional Mean Optical-flow (MDMO). The framework amplifies micro-expression motion information, significantly enhancing the recognizability of expression intervals in optical flow feature maps. By employing a sliding window localization method, the framework accurately captures the start and end frames of expressions, enabling effective merging and post-processing of expression intervals. This approach achieves an optimized balance between precision and recall while reducing the false positive rate. Our method achieved state-of-the-art results on the CAS(ME)<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> and SAMM-LV datasets and secured third place on the MEGC2024 testing set, demonstrating both the effectiveness and transferability capabilities of our approach. Our code is available at https://github.com/ceasorcaiIMEGC2024