Research And System Implementation Of Human Image Segmentation Based On Fully Convolutional Neural Network

Image segmentation is a research hotspot in the field of computer vision.Image segmentation technology is used to understand and segment complex scenes;mine the valuable information we need,and provide a research basis for other computer vision tasks.As a sub-problem of image segmentation,character image segmentation is also a research focus today,and has applications in security monitoring,photo character processing,and so on.However,due to the change of background and characters in the character image,the automatic segmentation of the character image faces huge challenges.Current character image segmentation is mainly divided into traditional character image segmentation and deep learning-based image segmentation methods.Traditional character image segmentation usually requires users to perform auxiliary operations to complete the image character segmentation.This is inefficient when processing a large number of pictures.The segmentation effect in complex scenes is usually not satisfactory,and automatic segmentation cannot be achieved.The deep learning-based character segmentation method uses a convolutional neural network to automatically learn and extract character information from the image,and trains the network by using massive labeled data to complete automatic segmentation of the character image.This paper makes an in-depth study on the fully convolutional neural network,and proposes an automatic character segmentation method,which is based on the improved character segmentation network structure of the fully convolutional neural network,so that it can achieve high-precision segmentation of person images.The main work and results of this article are as follows:(1)An improved automatic segmentation network C-FCN for people's images.This network is based on a fully convolutional neural network.It introduces hole convolution and spatial pyramid pooling in the encoding stage,and improves the upsampling in the decoding stage to improve the network accuracy and speed in segmentation.Finally,an experimental comparative analysis of the network is performed on the Baidu character image dataset.The segmentation results have a certain improvement over the fully convolutional neural network,which proves the effectiveness of the segmentation network proposed in this paper.(2)In this paper,the non-photorealistic rendering of the image is studied;and the image is stylized for ink painting.After realizing the segmentation of the character image,the foreground and background of the image are separated,and the stylized effect processing of the character image background is done separately.At the same time,based on the above work,this paper designs and implements an automatic processing system for character images,which can separately segment character images and stylize one-key processing of character image background ink.After the user uploads the image,without complicated manual interaction,one-click can realize the character image segmentation and background ink stylized processing.Finally,the system is tested and verified,and good results are obtained.The research in this paper has certain application value in real life.