Research On Image Semantic Segmentation Based On Deep Learning

Image segmentation has been a research focus in the field of computer vision,and deep learning is hot spot in recent stage with rapid development in artificial intelligence,so the deep learning applications suing vision techniques has become one of the most focus issues of researchers all around the world.The development of deep learning in the past ten years has made a very constructive breakthrough in the field of artificial intelligence.It is one of the most popular intelligent applications in the era of Internet and Big data environment at the present stage.Deep learning has achieved remarkable success in speech recognition,image retrieval,image content analysis,natural language processing,computer vision,video analysis,multimedia analysis and so many broad fields.As a result,a variety of computer vision problems are trying to use the deep learning methods for a more groundbreaking research.This paper focuses on the research of image semantic segmentation based on deep learning method.The first research is a learnable contextual regularization for semantic segmentation of indoor scene images.The second research is a semantic segmentation guide for conditional generative adversarial networks for better pixel level accuracy and more correlation between pixels.We first introduces the basis knowledge of traditional methods and deep learning method in image semantic segmentation tasks.As well as introduces the current most popular adversarial learning methods and the details of generative adversarial networks.In this paper,the advantages of the convolution neural network in image semantic segmentation task are analyzed,and the feasibility and advantages of using the generative adversarial networks for the task are discussed too.Semantic segmentation of indoor scene images has a wide range of applications.However,due to a large number of classes and uneven distribution in indoor scenes,mislabels are often made when facing small objects or boundary regions.Technically,contextual information may benefit for segmentation results,but has not yet been exploited sufficiently.In this paper,we propose a learnable contextual regularization model for enhancing the semantic segmentation results of color indoor scene images.This regularization model is combined with a deep convolutional segmentation network without significantly increasing the number of additional parameters.Our model,derived from the inherent contextual regularization on the indoor scene objects,benefits much from the learnable constraint layers bridging the lower layers and the higher layers in the deep convolutional network.The constraint layers are further integrated with a weighted L1-norm based contextual regularization between the neighboring pixels of RGB values to improve the segmentation results.Semantic segmentation is widely researched as a hot topic in computer vision.Remarkable results have been achieved based on deep convolutional neural networks(DCNN).Generative adversarial network(GANs)is one of the most important new models in the field of deep learning.Most of the existing researches mainly focus on the task of image generation and style transfer,the image semantic segmentation research is not enough in-depth.Typically,we ameliorate GANs further and propose conditional segmentation GAN(CSGAN)to tackle the issue of semantic segmentation.In our model,segmentation model as generator receives RGB images as inputs and outputs segmentation results.Besides a standard pixel-wise reconstruction loss,discriminator tries to distinguish labels from segmentation results.Discriminator provides a self-learning adversarial loss for global information,not dependent on manual loss item.Our CSGAN is end-to-end training,and results show that our CSGAN can not only improve the integrality of single object,but also maintain the independence between objects.Our experimental results on NYUDv2 indoor scene dataset demonstrate the effectiveness and efficiency of the proposed method.The first research has good results both on RGB and HHA images for effectively segmentational precision improve and more clear edge effect.We proposed conditional segmentation adversarial network has proved the effectiveness of segmentation network in adversarial learning model.We experimented on CamVid dataset for 2-class and 12-class segment tasks,our model is also end-to-end and help the correlation of pixels more closer?...