摘要
图像语义分割一直是计算机视觉中具有挑战性的任务之一。目前多数基于卷积神经网络的语义分割算法存在分割结果不精确,不连续等问题。为了提高图像分割效果,提出了基于生成对抗学习的图像语义分割网络模型。该模型由生成网络和判别网络两部分组成。生成网络含有五个模块,主要作用是生成语义分割图,判别网络与生成网络进行对抗训练,优化生成网络以使生成图像更加接近于Ground Truth。通过在Pascal VOC 2012数据集上对图像进行语义分割的分析,验证了该算法可以有效提高图像语义分割的精度。
Image semantic segmentation has always been one of the most challenging tasks in computer vision.At present,most semantic segmentation algorithms based on convolutional neural networks have the problem of inaccuracy and discontinuity of segmentation results.In order to improve image segmentation effect,propose image semantic segmentation network model based on generative adversarial learning,which consists of generating network and discriminating network.The generative network contains five modules,and the main function is to generate semantic segmentation image.The main role of the discriminative network is to confront with the generative network to make the generated semantic segmentation image closer to Ground truth.The semantic segmentation results on the Pascal VOC 2012 show that the proposed algorithm can effectively improve precision of image semantic segmentation.
作者
张嘉祺
赵晓丽
董晓亚
张翔
ZHANG Jiaqi;ZHAO Xiaoli;DONG Xiaoya;ZHANG Xiang(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《传感器与微系统》
CSCD
2019年第8期50-53,共4页
Transducer and Microsystem Technologies
基金
上海市科委资助项目(15590501300)
国家自然科学基金资助项目(61461021)
关键词
语义分割
生成对抗网络
深度学习
数字图像处理
semantic segmentation
generative adversarial network
deep-learning
digital image processing
