Unsupervised Deep Learning And Its Application In Image Segmentation

Intelligent society is the general trend.Image segmentation,as one of the basic work in the field of computer vision,has always been the focus and hot spot of research in related fields.Because of the strong fitting ability of neural network,many neural network image segmentation models based on supervised learning are proposed.In order to overcome the shortcomings of supervised learning,which requires a large number of labeled samples,and needs to train the model in advance,the unsupervised neural network segmentation model is gradually concerned.In order to solve the problem that it is difficult to divide the same object into one when the previous model fails to solve the problem that the same object has significant color difference,this paper first debugs SLIC algorithm and felzenszwal B super-pixel segmentation algorithm,and observes whether the super-pixel block obtained by different parameters can get the boundary of the object in the picture.Then a small convolutional neural network is trained by using the super pixels obtained by SLIC algorithm and felzenszwalb algorithm respectively.The first layer of the two networks is cut off,connected to the third network,the parameters of SLIC are modified,the training of the third network is completed,and the initial segmentation is obtained.Finally,the conditional random field is used for post-processing to get the final segmentation image.The experimental results show that the proposed unsupervised cascade segmentation model can effectively solve the problem that it is difficult to divide the same object into one when it has significant color differences.At the same time,the overall segmentation effect is good,and it can effectively extract the texture and color features of the object in the image.