Research And Application Of Intelligent Analysis Algorithm For Anime Characters

With the increase of computer computing power and the increase of data size,artificial intelligence has entered a new period of prosperity and development.The application of artificial intelligence involves all walks of life,such as face recognition,speech recognition,pedestrian detection and so on.At the same time,with the vigorous development of the animation industry,its audience is expanding,and the market for the animation industry has also greatly increased.In the anime industry,there are few applications for artificial intelligence,and the development space for combining artificial intelligence and anime is very large,such as the automatic generation of cartoon artwork covers,and the intelligent editing of anime clips.This will not only save a lot of manpower and material resources,but also make the use of anime as the core application more entertaining,attractive and competitive.In this thesis,we propose an intelligent character analysis algorithm based on target detection,and build a service application system based on intelligent analysis of anime characters.The main work of this thesis is as follows:(1)Aiming at intelligent analysis of anime characters,an algorithm based on target detection is proposed.It is based on SSD target detection algorithm and improved from three aspects.The first aspect is to use the network structure ResNet34 with more characteristic expression capabilities to replace the pre-network structure;The second aspect is to reset the shape of the default box based on the animated character's labeling features;The third aspect is to modify the loss weight of positive samples in the loss function to alleviate the imbalance problem.In terms of experiments,experiments were conducted on three major algorithms(Faster R-CNN?YOLO and SSD)to verify the superiority of our choice of basic algorithm SSD.Then,the training data was enhanced and experimented by three methods of rotation,cropping and color change,which verified the effectiveness of the data enhancement strategy.Finally,we conducted experiments on three aspects of the improvements we have made to verify its effectiveness.(2)Constructed an application system that takes the anime characters analysis capability service as the core.We have independently built a hierarchical server that is mainly used for algorithm training,and we have deployed a network service for the anime character intelligence analysis capability on this server.The hierarchical server is divided into:application layer,platform layer,and device layer,which is more convenient to manage.The deployment of network services is in the form of Nginx ?uWSGI + Django,with stable performance and good support for concurrent requests.Then it showed the overall flow of the application of the anime character intelligent analysis capability service.The service application has been released on the MIGU Loop APP and can be downloaded directly from the MIGU Loop website at http:/www.migudm.cn/.Finally,an example of the use of this capability service in the MIGU Loop APP is shown.