Research On Image Segmentation Method Based On Saliency Detection

Image segmentation is one of the key research topics in the field of computer vision,which is widely used in image recognition,detection and medical image processing.Because there are many kinds of image,and it is easy to be affected by light,shadow,uneven gray,complex background and other factors,it is difficult to segment the image.Although the image segmentation algorithm has made some progress in recent years,there is still much room for improvement in segmentation speed,accuracy,robustness and so on.This paper studies the segmentation of natural image based on saliency detection model.The main research contents are as follows:Aiming at the problem that the existing saliency detection method has blurry edges,the paper proposes an LBF model combined with the concept of multi-scale based on the LBF model.This method can solve the segmentation of severe grayscale uneven images and realize the image Accurate segmentation.In order to improve the segmentation efficiency,the algorithm uses a salient target detection mechanism to realize the automatic generation of the initial contour of the level set function,and then uses the multi-scale LBF model to solve the minimum value of the energy functional to obtain the optimal level set evolution of the target position.Experimental results show that the algorithm has a certain improvement in segmentation accuracy.In order to solve the problem of poor robustness and fuzzy segmentation boundary of existing methods of significance detection,this paper proposes an image segmentation model based on the significance of multi-scale depth features by using convolution neural network.The model is divided into three parts:in the first part,three deep convolution neural networks(CNN)are used to extract multi-scale features,which can obtain high-quality visual saliency map;in the second part,a guide filter is extended to a layer in the network,which can ensure the transmission of saliency information between pixels and restore the full resolution saliency map;in the third part,the guide filter layer is used to extract multi-scale features The level set function of the generated saliency map is learned to get the precise boundary of the saliency target and realize the segmentation of the image.Experiments on msra1 0k and sed2 datasets show that the performance of the proposed algorithm is improved in accuracy and recall.