Research On Image Segmentation Algorithm Based On Spectral Segmentation And Convolutional Network

Image segmentation is to classify pixels in an image,that is,to subdivide pixels with similar attributes into several disjoint regions.This paper studies image segmentation algo-rithms based on spectral segmentation and convolutional networks.The main contents are:Firstly,spectral clustering is a widely used unsupervised clustering algorithm that has a good clustering effect in many cases.However,for high-resolution images,the application is limited due to the high computational complexity.In this paper,we propose a fast spectral clustering algorithm based on wavelet basis decomposition.According to the hierarchical structure of wavelet decomposition,the algorithm reduces the dimension of the eigendecom-position of the graph Laplacian by wavelet basis matrix.The low-frequency eigenvectors of the entire graph Laplacian are solved hierarchically from wavelet subspaces with different levels.Its computational complexity is O(n)+u(m2),where n and m are the number of pixels and selected wavelet coefficients in an image,respectively.To verify the effectiveness and performance of the proposed algorithm,a series of experiments were done on both the Weizmann and BSDS500 datasets and find that our method,which in practice provides on av-erage about 5× speed-up to the eigendecomposition computation required for the Laplacian matrix with comparable segmentation accuracySecondly,the research on image semantic segmentation has entered into the field of deep learning,in which the convolutional network has achieved excellent results and shown strong applicability.This paper proposes two convolutional network models for image semantic segmentation.One uses PeleeNet as the network decoder to extract image features,and the decoder uses a combination of skip connection and maximum pooling index,which is called PNet.The other uses ENet as the backbone network,the convolution module and upsampling module of the ENet network are modified.Simultaneously,the dilated convolution modules are fused in turn,which is called LBNet.During the training process,a multi-resolution network strategy is utilized.By sending pictures of different resolutions to the network for training,the same image area is encouraged to use the same label at different resolutions to enhance the interaction between pixels.To a certain extent,the segmentation speed and precision of the model on the Cityscapes dataset are improved.Thirdly,image processing technology is increasingly applied in medical image diagno-sis.Therefore,this paper designs a cervical cell segmentation and lesion detection system by using the Caffe framework.This system uses the methods proposed in this paper to detect cervical cells,and can directly obtain image segmentation and lesion detection results by in-putting cell pictures.At the same time,the differences between the convolutional network and spectral segmentation algorithm in image segmentation tasks are further compared.