Handwriting skeleton refinement method, system and equipment based on deep reinforcement learning, and medium

The invention provides a handwriting skeleton refinement method based on deep reinforcement learning, and the skeleton refinement method is an end-to-end algorithm, and can effectively reduce the model complexity. The method comprises the following steps: acquiring a binary image of handwriting, and calculating contour pixel points of a signature; constructing a Markov decision process, constructing a neural network as a deep Q network DQN in deep reinforcement learning, inputting the binarized handwriting image as a state into the deep Q network, performing training according to the Markov decision process, and obtaining a binary handwriting image through each step of iteration. And guiding the intelligent agent to update the state in the iteration process by using the IoU values of the handwriting image and the skeleton before and after the action, and removing redundant pixels in the binary image to form the handwriting skeleton. The handwriting features and accuracy are improved, main structure informatio