The 2013 6th International Congress on Image and Signal Processing(CISP 2013)and the 2013 6th International Conference on BioMedical Engineering and Informatics(BMEI 2013)will be jointly held in Hangzhou,China.CISP-BM...The 2013 6th International Congress on Image and Signal Processing(CISP 2013)and the 2013 6th International Conference on BioMedical Engineering and Informatics(BMEI 2013)will be jointly held in Hangzhou,China.CISP-BMEI2013 is a premier international forum for scientists and researchers to present the state-of-the-art of multimedia,signal processing,biomedical engineering and informatics and to discuss future research challenges.Co-locating two conferences we aim to promote collaboration among multiple areas,especially the interactions of engineering,computing,and medicine.CISP’13-BMEI'13 is technically co-sponsored by the IEEE Engineering in Medicine and Biology Society.展开更多
大多数基于深度学习技术的视觉设备都配备了图像信号处理(Image Signal Processing, ISP)过程以实现RAW数据到RGB图像的转换,同时集成数据预处理过程以完成高效的图像处理。该文在同时考虑ISP和数据预处理过程影响基础上,搭建了深度学...大多数基于深度学习技术的视觉设备都配备了图像信号处理(Image Signal Processing, ISP)过程以实现RAW数据到RGB图像的转换,同时集成数据预处理过程以完成高效的图像处理。该文在同时考虑ISP和数据预处理过程影响基础上,搭建了深度学习视觉应用实验平台,提出针对ISP过程的图像缩放攻击,即精心制作对抗RAW经过ISP过程得到攻击图像,一旦缩放到特定尺寸就会呈现出完全不同的样貌。由于所提出的攻击是由基于ISP过程的梯度驱动的,因此构建了一个等效模型来学习目标ISP的转换过程,利用等效模型的近似梯度来发动攻击。该攻击平台的构建涵盖了深度学习算法、图像处理及对抗攻击优化等内容,有助于学生深入学习和理解基于深度学习视觉应用的任务处理原理以及深度学习模型的弱点,培养学生针对复杂算法问题的创新与实践能力。展开更多
文摘The 2013 6th International Congress on Image and Signal Processing(CISP 2013)and the 2013 6th International Conference on BioMedical Engineering and Informatics(BMEI 2013)will be jointly held in Hangzhou,China.CISP-BMEI2013 is a premier international forum for scientists and researchers to present the state-of-the-art of multimedia,signal processing,biomedical engineering and informatics and to discuss future research challenges.Co-locating two conferences we aim to promote collaboration among multiple areas,especially the interactions of engineering,computing,and medicine.CISP’13-BMEI'13 is technically co-sponsored by the IEEE Engineering in Medicine and Biology Society.
文摘大多数基于深度学习技术的视觉设备都配备了图像信号处理(Image Signal Processing, ISP)过程以实现RAW数据到RGB图像的转换,同时集成数据预处理过程以完成高效的图像处理。该文在同时考虑ISP和数据预处理过程影响基础上,搭建了深度学习视觉应用实验平台,提出针对ISP过程的图像缩放攻击,即精心制作对抗RAW经过ISP过程得到攻击图像,一旦缩放到特定尺寸就会呈现出完全不同的样貌。由于所提出的攻击是由基于ISP过程的梯度驱动的,因此构建了一个等效模型来学习目标ISP的转换过程,利用等效模型的近似梯度来发动攻击。该攻击平台的构建涵盖了深度学习算法、图像处理及对抗攻击优化等内容,有助于学生深入学习和理解基于深度学习视觉应用的任务处理原理以及深度学习模型的弱点,培养学生针对复杂算法问题的创新与实践能力。
