Research on infant Expression Recognition based on ResNet and Improved Intensity Normalization Algorithm

Infancy is an important human stage. Because babies' verbal ability is not yet perfect, we can only judge their emotions and states through facial expressions and body movements. At present, there is almost no infant expression data available for deep neural network training. Based on the above problems, we collects 1565 baby expression pictures from the Internet as a baby expression recognition data set. In addition, we use the deep neural network ResNet50 (Residual Network) as the infant expression recognition model, and preprocesses the original infant expression pictures, and proposes an improved intensity normalization algorithm. The experimental results show that compared with the traditional intensity normalization algorithm, the recognition rate of the improved intensity normalization algorithm can reach 87.16%, which improves the accuracy by nearly 5%.