【摘要】 图像语义分割致力于将图像中的内容识别出来,即识别出图像中每个位置像素的类别。基于全卷积神经网络的语义分割方法取得了良好的进展,然而这种方法需要大量的且极其耗时的像素级别的标注,为了解决这个问题,基于弱监督和半监督的研究逐渐受到关注。在目前的弱监督和半监督算法中,大部分使用基于手工设计的算法来生成图像区域建议,没有充分利用图像的边界框标注信息。针对这个问题,本文提出了基于学习的弱监督和半监督图像语义分割算法。在全卷积分割网络基础上,利用边界框标注信息,学习出一个通用的图像二元分割模型,再生成图像区域建议,可以更好的利用图像的全局信息和边界框的位置信息。在基准数据集Pascal VOC上的实验结果证明,本文的算法性能已经超过目前的优秀的算法。更多还原

【Abstract】 Image semantic segmentation is dedicated to recognizing the content of the image,and will recognize the category of the pixel at each position in the image. The semantic segmentation method based on Fully Convolutional Networks( FCN) has made great progress. However,this method requires a large number and extremely time-consuming pixel-level labels. In order to solve this problem,research based on weak supervision and semi-supervision has gradually attracted attention. How to make full use of bounding box labels is a very difficult problem. In the current weakly supervised and semi-supervised algorithms,most of them use handcrafted algorithms to generate image region proposals,which is relatively blunt and rough,and does not make full use of the image’s bounding box annotations. In response to the above problems,this paper proposes an image semantic segmentation algorithm based on weakly and semi-supervised learning. Based on the fully convolutional segmentation network,the bounding box is used to annotate information to learn a general image binary segmentation model,and then generate image region suggestions. This learningbased algorithm generates image region suggestions,which can make better use of the global information of the image and the position information of the bounding box. The experimental results on the benchmark dataset Pascal VOC mIoU 67.6 prove that the performance of the algorithm in this paper has surpassed the current best state-of-the-art algorithm.更多还原