• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文学位 >Real-time embedded algorithms for local tone mapping of high dynamic range images.
【24h】

Real-time embedded algorithms for local tone mapping of high dynamic range images.

机译：用于高动态范围图像的本地色调映射的实时嵌入式算法。

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Digital imaging is a widespread and powerful technology, and yet the quality of digital images suffer from a fundamental mismatch in the physical devices used in imaging systems: new technology camera sensors can capture scenes of much wider dynamic range than standard display devices can display. The result of this mismatch is that printed and displayed images are commonly lacking in clear detail. Cameras are able to capture the detail in, for example, both shaded and sunlit areas of a natural scene, but when the image is displayed either the sunlit area is saturated to become uniformly bright, or the shaded area is saturated to become uniformly dark. The problem is even more pronounced for synthetic images produced by fusing multiple views of a scene. Image processing is required to compress the dynamic range of the image before it is displayed, and to control its brightness and contrast. Ideally, this processing would be done in real-time, and embedded within the camera or display.; Published algorithms for processing images of wide dynamic range have been implemented mostly on workstations and applied to stored images, without hardware and timing constraints. The algorithms have been developed without taking into account the difficulties inherent in hardware implementation, and are not suitable for implementation in embedded systems that work in real time. This limits their application. This work proposes to develop algorithms for processing images of high dynamic range that are suitable for implementation on embedded systems, and for real-time applications.; One difficulty in developing real-time algorithms for image enhancement is the interdisciplinary nature of the work. Both image processing and digital hardware implementation aspects must be carefully considered, and these two aspects are intertwined so that they must be considered simultaneously. It is not sufficient to simply translate algorithms that perform well in floating-point mathematics on workstations into hardware; the hardware in a real-time application so constrains what can be done that the algorithm and the hardware must be developed hand-in-hand. The quality of the system as a whole depends heavily on both signal processing and hardware performance aspects.; The intellectual merit of the work lies in the following key concepts. Algorithms for local tone mapping of high dynamic range images that are uniquely appropriate for implementation on embedded systems and can do the processing in real-time are developed. Both hardware performance and signal processing aspects of the design are simultaneously considered so that system-level trade-offs can be made. The work also addresses an additional and crucial manifestation of computing implementations: the effect of fixed-point quantization and computational structure on the quality with which a digital signal processing algorithm is implemented.; The final outcome of our research is a local tone mapping operator, implemented on an FPGA that can process a 3x32-bit high dynamic range color image of 1024x768 pixels. It has a maximum operating frequency of 65.86 MHz, which is compatible with a video rate of 60 frames per second. The quality of the images processed by this operator was compared to a floating-point state-of-the-art local tone mapper considered to be a gold standard. Images processed by this operator had a peak signal-to-noise ratio of about 35dB on average with respect to the gold standard. The operator requires relatively simple hardware, and is well suited for embedded real-time application.
机译：数字成像是一项广泛而强大的技术，但是数字图像的质量却受到成像系统中所用物理设备的根本不匹配的困扰：新技术的摄像头传感器可以捕获比标准显示设备显示的动态范围大得多的场景。这种不匹配的结果是打印和显示的图像通常缺少清晰的细节。相机可以捕获自然场景中阴影和阳光照射区域中的细节，但是当显示图像时，阳光照射区域将饱和变为均匀明亮，或者阴影区域将饱和变为均匀黑暗。对于通过融合场景的多个视图而生成的合成图像，该问题甚至更为明显。需要进行图像处理以在显示图像之前压缩其动态范围，并控制其亮度和对比度。理想情况下，该处理将实时完成，并嵌入到摄像机或显示器内。已发布的用于处理宽动态范围图像的算法大部分已在工作站上实现，并已应用于存储的图像，而没有硬件和时序限制。这些算法的开发没有考虑到硬件实现中固有的困难，因此不适合在实时工作的嵌入式系统中实现。这限制了它们的应用。这项工作提出了开发用于处理高动态范围图像的算法，该算法适合在嵌入式系统上实现，并且适合实时应用。开发用于图像增强的实时算法的困难之一是工作的跨学科性质。必须仔细考虑图像处理和数字硬件实现方面，并且这两个方面是交织在一起的，因此必须同时考虑它们。仅仅将在工作站上的浮点数学运算中表现良好的算法简单地转换为硬件是不够的。实时应用中的硬件，因此限制了必须完成的工作，算法和硬件必须齐头并进。整个系统的质量在很大程度上取决于信号处理和硬件性能方面。该作品的知识价值在于以下关键概念。开发了用于高动态范围图像的局部色调映射的算法，该算法非常适合在嵌入式系统上实现，并且可以实时进行处理。同时考虑了设计的硬件性能和信号处理方面，以便可以进行系统级的权衡。该工作还解决了计算实现的另一个关键性体现：定点量化和计算结构对实现数字信号处理算法质量的影响。我们研究的最终结果是在FPGA上实现的本地色调映射运算符，该运算符可处理1024x768像素的3x32位高动态范围彩色图像。它的最大工作频率为65.86 MHz，与每秒60帧的视频速率兼容。将由该操作员处理的图像质量与被认为是金标准的浮点最新的本地色调映射器进行了比较。相对于金标准，由该操作员处理的图像平均具有约35dB的峰值信噪比。运营商需要相对简单的硬件，非常适合嵌入式实时应用。

著录项

  • 作者

    Hassan, Firas.;

  • 作者单位

    The University of Akron.;

  • 授予单位 The University of Akron.;
  • 学科 Engineering Electronics and Electrical.
  • 学位 Ph.D.
  • 年度 2007
  • 页码 134 p.
  • 总页数 134
  • 原文格式 PDF
  • 正文语种 eng
  • 中图分类 无线电电子学、电信技术;
  • 关键词

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号