Cartoon Character Recognition Based On Mask R-CNN

Animation industry is an important part of today’s cultural and creative industry,forming a huge consumer market around the world,and even becoming a pillar industry in developed countries.However,in China,the animation industry has just emerged in the past 10 years,with relatively backward development,but it has shown strong vitality and vigorous development momentum.How to promote the development of China’s animation industry,promote the combination of animation and science and technology,and promote Chinese animation to go out is of great significance.In recent years,object detection in computer vision has been widely concerned by many researchers at home and abroad because of its large application requirements in the real world,such as automatic driving,video surveillance,robot vision,etc.In the field of computer vision,image recognition has developed by leaps and bounds in recent years,and image recognition has become a research hotspot.This paper hopes to make a cartoon character recognition system to facilitate people’s search of cartoon characters and their works in daily life,so as to increase the number of users and activity in the animation market and promote the development of China’s animation industry.In view of the above requirements,the main work of this paper is as follows:(1)Data set production:download pictures and intercept video pictures through network search to obtain the original picture data,then expand the picture data to obtain the required image data set,frame the people in the picture with the annotation tool,and generate the annotation information of each picture.Use user-defined script to randomly divide the data set for model training.(2)Use different feature extraction networks to extract feature,and improve the original Faster R-CNN algorithm:Integrate Feature Pyramid Networks and replace ROI Pooling by USING ROI Align Mask R-CNN to improve the network,and compare and analyze the identification results of various types of people.The improved model based on feature pyramid network solves the problem of large scale difference of people and difficulty in identifying small targets,and has a great improvement in recognition accuracy.The prototype system of cartoon character recognition is made by the above methods.