Answers are grounded in this paper and cite their sources.
Micro-expression (ME) is a form of reflexive behaviour and indirect communication in which an individual expresses their emotions via facial muscle movements.Due to the nature of ME that only appears for a fraction of a second, and even a trained individual will have a hard time detecting and recognizing it.In order to avoid human error and achieve better results, an automatic ME recognition system is introduced.In this work, a transfer learning approach is utilized to recognize the static facial microexpression images.Six pre-trained convolutional neural network (CNN) models, including the AlexNet, SqueezeNet, GoogleNet, EfficientNet-b0, ResNet 50, and MobileNet-v2 are employed and evaluated on the improved version of the Chinese Academy of Sciences Micro-expression (CASME II) database.The pre-trained CNN models are compared to each other by their accuracies.We conducted the experiments under three different settings, in Setting 1 the parameter of entire learnable layers are unaltered, setting 2 is by freezing the first 20% of the learnable layers and Setting 3 the first 50% of the learnable layers are frozen.AlexNet obtained higher accuracy of 99.84% in Setting 3. The freezing learnable layer approach is able to improve a pre-trained model's accuracy and accelerate training time by not altering the parameters of the frozen layers.This research not only benefits the psychology field, but it also benefits the marketing field, security purposes, and other applicable fields.
No takes yet. Share an insight, caveat, or question.
A 2024 study studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: