Exploring the Caricature Style Identification and Classification Using Convolutional Neural Network and Human-Machine Interaction Under Artificial Intelligence

The objectives are to explore the application effect of deep learning (DL), neural network (NN) and human-machine interaction (HMI) technology in caricature style recognition and classification in the environment of artificial intelligence (AI). It provides a realistic basis for the integration of caricature and AI technology. The convolutional neural network (CNN) model is optimized, and the ARtoolkit, JAVA-based processing and photoshop are applied, combined with an augmented reality (AR) editor to register the logo of caricature. The results indicate that the output value of the optimized CNN is the largest near the fourth neural node, and the output values of the remaining neural nodes are almost all close to zero. The label value of the cat image is the same as the output value, and the image label range is [0, 10]. The use of AR technology can make caricature images have animation functions, giving people the visual experience that the lotus can move, and the caricature style has changed. When the screen is not touched, the fish under the lotus is in a static state. When the screen is touched with the hand, the fish swims away quickly, indicating that the function of HMI is realized. Therefore, the CNN and HMI technology under AI can successfully perform caricature style recognition and classification, which can provide an experimental reference for the subsequent intelligent development of caricature.