Image Semantic Segmentation Based On Deep Learning

In recent years,with the society stepping into the information age,constantly showing massive image data.Large scale image data in meet the demand of people’s entertainment and promote the development of society and also at the same time make image analysis and understanding becomes a problem that can not be ignored and solved urgently.Semantic image segmentation is considered to be one of the most important processing techniques in computer vision,and also is a key step of image analysis and scene understanding.Image semantic segmentation aims to classify each pixel of the image,and segment the image into several visually meaningful or interesting areas,so as to facilitate the subsequent image analysis and understanding.Noise may be arisen in the capturing and transmission process of the image,which not only corrupts true information of an image,but also seriously affects the visual effects of the image.Therefore,noisy image semantic segmentation becomes one of the most challenging problems in image analysis.The most traditional image segmentation methods are to extract the characteristics of the image itself,the segmentation process are more complex,and the segmentation results need to be further improved.With the rapid development of deep learning,image semantic segmentation has entered a new stage of development.Deep learning technology includes many kinds of neural network structures.Among them,the most commonly used deep convolutional neural network(DCNN)extracts the image features are of great advantage,is a very effective technique in machine learning.Therefore,how to use deep learning to improve the performance of image semantic segmentation has become a current focus of research.In order to improve the segmentation performance of the noisy image,this paper studies the image semantic segmentation algorithms based on deep learning at home and abroad,the semantic segmentation algorithms based on deep learning are sorted out and summarized.On the base of analyzing Fully Convolutional Network(FCN),we propose an improved FCN model(IFCN)for noisy image semantic segmentation by using the existing deep convolution neural network(Alexnet).The algorithm uses a new median pooling method instead of max pooling in the convolutional neural network,which can remove noise and preserve more boundaries information.Our model trains the whole deep network by a direct way of an end-to-end,pixels-to-pixels mapping with the back propagation algorithm.The proposed algorithm was run on PASCAL VOC2012 and Sift Flow datasets separately.The different network model with different parameters was trained for different types of noise images,including salt and pepper noise and Gaussian noise.Experimental results demonstrate that better semantic segmentation results can be obtained while removing noise at the same time.By comparing with the classical algorithm,the validity of this method is proved.