Study On Unsupervised Active Contour Model For Image Segmentation

Image segmentation is an important part of image analysis,and it is also an important task in computer vision.It can separate the object from the background and provide technical support for post-image processing,such as fault recognition,target recognition,change detection,regional positioning,etcImage segmentation is of great importance in computer vision.Among them,the active contour model has been widely used,because it can provide a smooth and closed boundary contour as a segmentation result.However,when attempting to segment an image with a large noise,the image is easily contaminated and the segmentation effect is poor.Existing methods assume that the pixels of each region are independent when calculating the energy function.This basic assumption makes the contour motion sensitive to noise.In addition,the implementation of the hierarchical set method is complex and time consuming,which limits its application in large image databases.Most of the current level set methods are calculated on the full image domain.However,the calculation of pixels away from the evolution contour has no meaning for acquiring the target boundary,resulting in an increase in complexity.Therefore,this thesis proposes an unsupervised active contour model based on mixed energy and Fisher criterion for image segmentation.In order to achieve accurate and efficient image segmentation,the local region information is transformed,and the Fisher criterion based on the global region information and local information is introduced.This approach optimizes the active contour model.In addition,by using the intra-class variance of the target area and the background area pixel as the energy weight of the target area and the background area.The ratio of the regional energy can be adaptively adjusted to improve the segmentation accuracy of the model.On this basis,the active contour model is improved.The energy weight is calculated by an internal variance.That is,the pixel difference between the target area and the background area is very robust for adjusting the area energy.The proposed method does not require prior knowledge and human intervention.Finally,the image segmentation accuracy is greatly improved.Experimental results show that this method has high segmentation precision and competitiveness without semantic priori.