Adaptive Enhanced Micro-expression Spotting Network Based on Multi-stage Features Extraction

Abstract

AbstractWhen micro-expressions are mixed with normal or macro-expressions, it becomes increasingly challenging to spot them in long videos. Aiming at the specific time prior of Micro-expression (ME)s, a ME spotting network called AEM-Net (Adaptive Enhanced ME Detection Network) is proposed. The network improves the spotting performance in the following four aspects. First, the multi-stage channel feature extraction module is proposed to extract feature information of different depths. Then, an attention spatial-temporal module was used to obtain salient and discriminative micro-expression segments while suppressing the generation of excessively long or short suggestions. Thirdly, a ME-NMS (Non-Maximum Suppression) network is developed to reduce redundancy and decision errors. Finally, two spotting mechanisms named anchor_based and anchor_free are combined in our method. Extensive experiments have done on prevalent \(\mathrm CAS(ME)^2\) and the SAMM-Long ME databases to evaluate the spotting performance. The results show that the AEM-Net achieves an impressive performance, which outperforms other state-of-the-art methods.KeywordsMicro-expressionSpottingAttention mechanismSpatial-temporal features