Research On Deep Learning Image Segmentation Model Based On Rough Calibration

Image Segmentation is to accurately separate the region of interest from other regions of the image,which is an important part of image processing,and the important tool of computer-aided diagnosis and anatomical structure research in medical treatment.Accurate and effective segmentation of i mages has always been a research hotspot,and it also provides strong support for subsequent image processing such as image detection and image recognition.Therefore,we not only need to study how to accurately segment the target from the image,but also need to study how to effectively segment in detail.In response to the above problems,this dissertation mainly does the following research work:First of all,this dissertation mainly studies the deep learning image segmentation U-net model based on rough calibration.We roughly circled the location of the target,and trains the U-net network.After that,we use the U-net model to roughly segment the image,and the other U-net model is used to further finely divide the above results.Through comparison experiments,it was verified that the U-net model based on coarse calibration had better segmentation effect than the U-net model,and the model achieved a better segmentation boundary.Secondly,in order to obtain better segmentation r esults in detail,this dissertation takes the segmentation results based on the coarse calibration U-net model as the initial segmentation,and combines with the level set method to propose the image segmentation model I.Then this dissertation uses gradient descent to solve the model I,and uses the finite difference method to numerically discretize the model I.The validity of the model is verified through numerical experiments.Based on the coarse calibration U-net model,the model obtained better segmentation details.Finally,this dissertation takes the segmentation results of the U-net model based on rough calibration as the initial segmentation and shape prior,and combines the level set method to propose the image s egmentation model II.We also used the finite difference method to obtain the numerical solution of the model,and realized the model II through programming.Then we compare the experimental results with the experimental results of the CV model and distance regularization level set method to prove the effectiveness of the model and obtain a better segmentation effect than model I.Therefore,we combined the deep learning image segmentation model and the level set method to further improve the model segmentation accuracy and achieve better results in details.