遥感影像数据与地理信息系统(geographic information system,GIS)矢量数据的配准是遥感与GIS集成的基础。目前遥感影像与矢量数据的配准关键在于遥感影像特征的提取,而现有遥感影像特征提取方法存在特征提取不完整、配准失败和精度不...遥感影像数据与地理信息系统(geographic information system,GIS)矢量数据的配准是遥感与GIS集成的基础。目前遥感影像与矢量数据的配准关键在于遥感影像特征的提取,而现有遥感影像特征提取方法存在特征提取不完整、配准失败和精度不高等问题。由此提出了一种基于Mask R-CNN(region-based convolutional neural network)的遥感影像与矢量数据配准方法,首先,利用Mask R-CNN模型提取影像的道路交叉口作为影像控制点;然后,依据几何拓扑关系筛选矢量数据道路交叉口作为矢量控制点,再根据遥感影像与矢量数据控制点的欧氏距离确定同名控制点;最后,以同名控制点为基础实现遥感影像与矢量数据的配准。选取上海市矢量数据和高分二号影像数据进行配准实验,实验结果表明,所提方法鲁棒性强、精度高。展开更多
In this paper,we suggest an adaptive watermarking method to improve both transparence and robustness of quantization index modulation(QIM) scheme. Instead of a fixed quantization step-size,we apply a step-size adapted...In this paper,we suggest an adaptive watermarking method to improve both transparence and robustness of quantization index modulation(QIM) scheme. Instead of a fixed quantization step-size,we apply a step-size adapted to image content in each 8×8 block to make a balance of robust extraction and transparent embedding.The modified step-size is determined by contrast masking thresholds of Watson’s perceptual model.From a normalized crossed-correlation value between the original watermark and the detected watermark,we could observe that our method is robust to attacks of additive white Gaussian noise(AWGN),Salt and Pepper noise and Joint Photographic Experts Group(JPEG) compression than the original QIM.By taking into account the contrast insensitivity and visible thresholds of human visual system,the suggested improvement achieves a maximum embedding strength and an appropriate quantization step-size which is consistent with local values of a host signal.展开更多
基金supports of China NNSF(Grant No.60472063. 60325310)GDNSF/GDCNLF(04020074/ CN200402)
文摘In this paper,we suggest an adaptive watermarking method to improve both transparence and robustness of quantization index modulation(QIM) scheme. Instead of a fixed quantization step-size,we apply a step-size adapted to image content in each 8×8 block to make a balance of robust extraction and transparent embedding.The modified step-size is determined by contrast masking thresholds of Watson’s perceptual model.From a normalized crossed-correlation value between the original watermark and the detected watermark,we could observe that our method is robust to attacks of additive white Gaussian noise(AWGN),Salt and Pepper noise and Joint Photographic Experts Group(JPEG) compression than the original QIM.By taking into account the contrast insensitivity and visible thresholds of human visual system,the suggested improvement achieves a maximum embedding strength and an appropriate quantization step-size which is consistent with local values of a host signal.
