【中文摘要】随着数字图像处理技术的高速发展,图像处理的应用已经深入到国计民生的方方面面。图像增强作为图像处理中的一个方向有着重要的意义,Retinex方法作为一种用于视觉场景再现的一种近似最优的方法正在迅速的为人们所使用, Retinex算法是一种能够改善图像亮度、对比度、清晰度的一种非线性变换,它能同时提供图像的动态范围压缩、色觉恒常、色彩再现三大优点,已经成功应用于各种静态图像的处理,然而由于图像处理所涉及的数据量比较庞大,图像处理速度成为了制约其发展的瓶颈之一,实时Retinex图像处理技术正是在这种需求下应运而生。图像处理硬件已经得到了飞速的发展,常见的专用的图像处理芯片价格一般较高,灵活性差,在扩展能力上也有所,我们很难在上面实现新的功能和算法。TI公司生产的TMS320DMDM2数字信号处理器(DSP)是当前业界公认的有着良好性能的多媒体处理器,它是一款专门面向多媒体应用而设计开发的32位定点DSP芯片,它基于CxDSP内核,采用了两级缓存机制并且提供了极为丰富的外设接口,可以很方便的实现图像的实时处理,用于多媒体应用开发,在硬件实现上也比较方便,同时TI公司还提供了功能强大的开发软件和丰富的库函数和API函数,所以,该芯片成为目前进行实时图像处理的理想平台之一,为实时图像处理提供了良好的基础。本文首先在介绍了几种经典的图像增强方法的基础上引入了Retinex方法,然后提出了一种新的基于双边滤波的类Retinex方法,同时分析并比较了Retinex发展中的各种经典方法和本文提出的类Retinex图像增强方法,并通过定量指标评价了这几种经典算法和本文算法在提高图像质量、颜色保持以及计算速度方面的优劣情况,分析了他们的优缺点,实验表明Retinex方法对灰度图像和彩色图像效果明显,证实了本文方法不仅可以实现图像的增强效果,而且可以提高处理速度,为实时处理提供了可能。接着本文以合众达公司的SEED-VPM2平台为例,实现并优化了Retinex算法,最后,文章指出了课题以后的研究方向,TI公司推出的达芬奇数字媒体处理器将成为更好的平台。
【英文摘要】As the rapid development of image processing technology recent years,the application has been used to the national economy and the people’s livelihood deeply.Image enhancement plays a very important role as a direction of image processing,Retinex algorithm has been rapidly used by people as a kind of approximate optimal method used for scene reappearance,it is a nonlinear transform which can improve image brightness,contrast,resolution,it can also provide three characteristics:dynamic range compression,color constancy and color rendition.this method has been applied in all kinds of static images successfully,However,because image processing involved enormous amount of data,processing speed of image processing has became one of the bottle-necks restricting its development.Real-time image processing emerged as the times require.Hardware of image processing has been developed rapidly,usually special image processing chip has a high price,also it has a bad flexibility and a limit to expand so that it is hard to realize new functions and algorithms.Digital signal processor DM2 is currently recognized a well-behaved multimedia processor in the industry cycle, it is a 32-bit fixed-point DSP chip specialize for multimedia.It is based on Cx DSP kernel,using a two-level buffer mechanism and provides abundant peripherals interface so that it can be very convenient to realize real-time processing of image and it can be used to development of multimedia application,also the hardware implementation is more convenient,at the same time,TI provides a powerful development software,abundant library functions and API functions,so nowdays the chip becomes one of ideal platforms for real-time image processing and lays a solid foundation for real-time image processing.Firstly,this paper introduces Retinex method on the basis of several classic image enhancement methods,and then proposes a new comprehensive Retinex method based on bilateral filter,and then analyses and compares various traditional Retinex methods with method of this paper,also evaluation in improving image quality,color maintaining and computing speed was done,at the same time we analysis advantages and disadvantages of them,Experimental results demonstrate that Retinex method works well for gray images and color images and confirm that this new Retinex methods proposed can not only realize image enhancement but also accelerate image processing which provides the possibility for real-time processing.After that,this paper realize classical Retinex algorithms on SEED-VPM2 made by SEED and optimize the implementation of Retinex methods,Finally,the thesis points out future research directions,The article points out that DaVinci digital media processor launched by TI company will become better platform for choice.
【关键词】Retinex 图像增强 DM2 视频处理
【英文关键词】Retinex image enhancement DM2 video processing
【目录】Retinex图像增强技术研究及其DSP实现摘要3-4Abstract4-5第一章 绪论9-131.1 论文的选题背景9-101.2 研究目的和意义10-111.3 本文的主要工作11-13第二章 Retinex图像增强技术13-402.1 图像颜色空间13-142.1.1 RGB颜色空间132.1.2 YUV颜色空问13-142.1.3 HSI、HSV颜色空间142.2 经典的图像增强方法14-182.2.1 直方图均衡化15-162.2.2 同态滤波162.2.3 小波变换16-172.2.4 线性拉伸172.2.5 伽马矫正17-182.3 Retinex基本原理18-202.3.1 色觉恒常18-192.3.2 Retinex算法简介及其核心思想19-202.4 Retinex的发展20-292.4.1 随机游走算法212.4.2 循环算法21-232.4.3 中心环绕Retinex算法23-262.4.4 变分Retinex方法26-292.5 基于双边滤波的带有变增益偏差的类Retinex方法29-332.5.1 双边滤波29-312.5.2 改进的伽马矫正31-332.6 实验结果比较和分析33-392.6.1 图像质量评价33-352.6.2 图像结果分析35-392.7 结论以及未来研究的建议39-40第三章 基于DSP的Retinex实时图像处理40-603.1 DSP概论403.2 DSP芯片的基本结构和特点40-423.2.1 改进的哈佛结构413.2.2 多总线结构413.2.3 流水线操作41-423.2.4 专用的硬件乘法器423.2.5 特殊的DSP指令集423.3 多媒体数字信号处理器DM242-473.3.1 DM2的CPU43-443.3.2 DM2的存储系统44-463.3.3 DM2的外设46-473.4 DSP软硬件开发平台47-543.4.1 硬件平台SEED-VPM247-523.4.2 DSP的软件开发平台52-543.5 基于Retinex算法的DSP实现54-603.5.1 软件设计54-553.5.2 实验流程55-563.5.3 实验中需要考虑的问题56-573.5.4 软件实现57-583.5.5 实验结果58-60第四章 基于DSP的Retinex方法实现的优化60-694.1 基于结构的优化60-634.1.1 图像的块处理方法60-614.1.2 充分利用DM2的EDMA以及‘ping-pong’机制61-634.2 基于算法和软件的优化63-684.2.1 基于算法的优化63-654.2.2 基于软件的优化65-684.3 优化结果评估68-69第五章 总结与展望69-715.1 全文工作总结69-705.2 工作展望70-71参考文献71-75致谢75-76攻读学位期间发表的学术论文76