Abstract: Existing object region segmentation approaches based on deep convolutional neural networks are inefficient for object boundary segmentation. To address this problem, we propose an object region segmentation algorithm based on pixel level features and semantic level features. Firstly, we extract and get image＇s semantic information by deep convolutional neural networks with the attention module and get the coarse segmentation results through a specific pixel level classifier. Then, we exploit conditional random fields to model pixel level correlation, thus getting image＇s local features. Finally, we utilize an optimization framework to fuse the local detail features and global semantic features. Compared with other methods, the experimental results show that our method can improve the segmentation accuracy of boundary and be beneficial for object region segmentation.