Semantic Segmentation For Mulitple Objects Based On Visuial Structure Model

Semantic segmentation is a fundamental problem in the area of computer vision,and it is one of the great challenges in computer vision.The techniques on semantic segmentation lay the foundations for high-level vision tasks,while they have widely used in a large number of industrial applications.Recently,deep convolutional neural networks(DCNNs)have achieved great success in most of the computer vision tasks.Such as image classification,object detection,semantic segmentation and edge detection and so on.This huge success has received much high attention due to the efficient feature extraction methods of DCNNs.DCNNs have strong ability to capture high-level visual feature for visual tasks,and this motivates exploring the use of DCNNs for pixel-level labeling problems.However,in terms of semantic segmentation,it is still difficult for developing a universal segmented architecture to deal with multiple applications in different areas simultaneously.This paper focuses on how to solve the difficult and complex problems in semantic segmentation.We show that several efficient,trained end-to-end,pixels-to-pixels segmentation algorithms achieved high-performance results and applied in multiple areas.The main researches and contributions of this paper are as follows:To achieve the efficient inference algorithms for densely pixel-level classification tasks,we propose a Context-aware Network(CNet),which utilizes robust context information to improve segmentation results.CNet has two significant components: 1)a feature collection module(FCM)and 2)a novel layer named Res Gate layer,which was developed to select robust context information from the FCM.The two combined components explore context information to improve boundary segmentation accuracy.To describe the multiple objects and complex scene,we proposed the visual structure model conception including visual construction,visual learning and visual inference.We deployed the proposed visual structure model to different modules in our segmentation system.To develop a universal segmentation system,we proposed a novel deep learning based system for multi-object semantic segmentation tasks for different situations or applications.Such as common object segmentation,urban street scenes understanding,pedestrian segmentation and medical image segmentation.