【作者】 林财伟；

【导师】 杨季文；

【作者基本信息】 苏州大学 ， 计算机应用技术， 2011， 硕士

【摘要】 随着数字图像的广泛应用,对图像的真实性、权威性做出认证显得越来越重要。研究人员提出了许多有效的图像认证算法,但仍然面临许多挑战,其中一个问题是图像认证的处理效率日益成为应用中的主要瓶颈。当前常用的加速方案主要采用可编程芯片作为协处理器,但这些方法在通用性等方面的局限性使其得不到广泛的应用。CUDA平台的推出为GPU在通用计算领域发挥作用提供了良好的软硬件平台。本文针对数字图像认证方法及其加速方案,探讨了CUDA平台下图像认证方法的设计与实现,具体工作包括以下几个方面:首先,分析了当前图像认证的主要方法与特点,探讨了引入CUDA平台提高处理效率的可行性与优越性。当前许多认证算法往往需要引入图像变换、人类视觉系统等模型,其执行过程具有较大的计算量和较高的可并行性,因此将基于CUDA平台的并行技术用于图像认证中能够有效提高算法的执行效率。其次,设计了一种基于CUDA平台的并行化的DCT域图像认证方法。将图像的分块变换与认证处理任务分配到GPU中并行处理,通过合理的任务分割与程序优化,充分利用了GPU的并行计算能力,获得了较好的加速效果。再次,设计了一种基于CUDA平台的并行化的DWT域图像认证方法。分析了Haar小波变换的计算特点,采用稀疏矩阵相乘的策略对其进行并行化。认证处理中结合人类视觉系统,自适应地嵌入图像特征水印。通过计算任务的并行化使得算法的处理效率得到有效的提高。最后,给出了CPU-GPU异构多核平台下图像认证系统的基本实施框架。探讨了CPU-GPU异构多核平台与多GPU调度方法。本文将CUDA平台有效地应用于图像认证领域,提高了图像认证过程中的处理效率,为图像认证的算法研究与并行方案提供了新的思路,具有一定的参考价值。更多还原

【Abstract】 Digital images are widely used in our life, and image authentication is becoming moreand more important. After years of research, researchers proposed many effiective image au-thentication algorithms, but there are still many challenges in this field. This paper presentsa new acceleration scheme for image authentication using CUDA, and the main researchworks are as follows:Firstly, the paper discusses the main image authentication method nowadays. Manyuseful concepts, such as signal transformation and HVS, are employed to improve the per-formance. Thus parallelizing these processes with CUDA will accelerate the performanceof the algorithm considerably.Secondly, the paper proposes a CUDA-based implementation of an image authentica-tion algorithm with GPU devices. The algorithm parallelizes the process of block-wise DCTand watermarking embedding. Experimental results show that the CUDA-based methodgains a good performance both in authentication results and operation speedup.Thirdly, a DWT-based image authentication algorithm and its parallel scheme are pre-sented, using sparse matrix multiplication strategy for the Haar wavelet transformation andimprove the efficiency of the process of image authentication.Finally, a basic framework for image authentication system is presented. Hybrid CPU-GPU platforms and multi-GPU are employed in the system.This paper applies CUDA platform to the field of image authentication and makes theprocess of image authentication more effiectively. The schemes provide new ideas to accel-eration of image processing, and have certain practical significance.更多还原