Research And Application Of Image Segmentation Algorithm Based On Fully Convolutional Networks

In recent years,as the level of science and technology has gradually improved,the proportion of image information in daily life has gradually increased,the image processing is getting more and more attention.Image segmentation occupies an important position in the image processing field and is a prerequisite for understanding image content,segmentation results have a direct influence on subsequent operations.Image segmentation applications can now be found in many fields,especially in the field of medical image.And image segmentation helps the medical diagnosis of the patient's location and can reduce the stress of the doctor in the course of treatment.But medical images are easily affected by external noise and the imaging process of medical images can produce additional defects due to instrument problems.Therefore,it is very important to design an algorithm that can improve the effect of medical image segmentation.At present,image segmentation has not found any segmentation method that can satisfy any image.Generally speaking,the idea of the classical image segmentation method is directly related to some characteristic of the image and is easily restricted by the image itself.Recently,combining with some fields of technology has become a major trend in the development of image segmentation technology.Artificial neural networks are now attracting the attention of many researchers,the proposed image segmentation algorithm based on fully convolutional network also opens up a new direction for image segmentation,however,the results of image segmentation using this method are not accurate enough and space information between pixels is not taken into account during the segmentation process.This paper will start from these two aspects and propose an improved algorithm based on fully convolutional neural network.The algorithm effectively improves the accuracy of fully convolutional network and optimizes its segmentation results.The main research results of this paper are as follows:1.In view of the problem that the image is not fine after the fully convolutional network segmentation,in this paper,we transform the energy function of graph cut algorithm to optimize the result.Experiments show that the algorithm effectively improves the accuracy of the final result of medical images.2.Image segmentation algorithm based on fully convolutional network lacks spatial consistency.Therefore,in this paper,the texture features of the image are introduced by Gabor filter in the segmentation process and use random forest to classify pixels again with the results obtained by graph cut algorithm.The results show that the algorithm can further improve the segmentation effect.3.This paper designs and implements a system of medical image segmentation based on improved algorithm,completes the system requirements analysis,functional module design and database design,finally introduces the system implementation process and the encoding of the system of the final test.