项目名称: 基于深度视觉注意机制的可逆水印及评价方法
项目编号: No.61303220
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 安玲玲
作者单位: 西安电子科技大学
项目金额: 25万元
中文摘要: 作为数字图像版权保护的有效手段,可逆水印技术能够无损恢复宿主图像内容,最大限度地满足医学、遥感等领域对内容无损的要求,具有更为广泛的应用前景。然而,随着盗版技术与手段的发展,保护图像显著内容、增强水印鲁棒性及公正评价2D水印成为可逆水印亟待解决的难题。为此,本项目拟结合人眼的视觉注意特性,从水印嵌入区域构造、水印嵌入与提取以及水印评价等三个环节出发,研究基于深度视觉注意机制的可逆水印及评价方法。首先,利用深度视觉显著模型实现频域水印嵌入区域的构造;其次,基于深度显著性约束进行多尺度特征点的优化筛选,在此基础上,设计几何不变域的双效零水印以增强水印抵抗常规图像处理与几何攻击的鲁棒性;最后,结合深度学习与支撑向量回归设计2D水印相似性度量测度实现2D水印的评价,从而为版权鉴别提供公正合理的辨识依据。本项目的研究将进一步完善可逆水印的理论框架,为数字图像版权保护提供理论依据与技术支撑。
中文关键词: 可逆水印;可逆信息隐藏;特征区域;鲁棒性;
英文摘要: As an effective way to protect copyright for digital image, reversible watermarking can losslessly recover the content of host image, and satisfies the requirement of content losslessness in the field of medical, remote sensing, and so on, showing wide application prospects. However, with the development of piracy techniques and skills, reversible watermarking is challenged to protect the sailent image content, improve the robustness and appropriately assess the 2D watermark. Therefore, this project exploits the visual attention characteristics of human eyes, and thus studies the reversible watermarking and assessment based on deep visual attention mechanism from three aspects which include region construction for watermark embedding, watermark embedding and extraction, and 2D watermark assessment. Specifically, a deep visual attention model is first used to construct the watermark embedding regions in the frequency domain. Secondly, the multi-scale feature points are optimally selected with a deep sailency constraint. Following this, a double effect zero-watermarking algorithm is designed via geometry invariant regions, which improves the robustness against the common image processing operations and geometric attacks. Finally, by combining the deep learning and support vector regression techniques, a similarity
英文关键词: Reversible watermarking;Reversible information hiding;Feature regions;Robustness;