【作者】 李艳梅；

【导师】 陈雷霆；

【作者基本信息】 电子科技大学 ， 计算机应用技术， 2013， 博士

【摘要】 图像增强是指以满足特定应用需求为目的，突出图像中感兴趣区域信息，抑制或去除其他信息，针对不同的应用而异的图像分析识别预处理，其目标是变换原图像信息为更加适合人机辨识的系列方法。对图像质量的要求也随着多媒体技术和产品的不断发展和在各领域的广泛应用而不断提高。而通常图像在获取过程中受成像设备、场景动态范围、光照条件等因素影响，使得图像质量下降，甚至于影响后续的人机图像分析识别。为此，需要对图像进行动态范围调整、对比度增强、彩色图像增强处理及视觉感知一致性等方面的处理来获得高质量的图像。论文以具体应用目标要求为基准，通过对图像的视觉效果增强相关理论和技术方法的深入研究，分析其在实际应用中存在的问题和缺点，进一步提出相应的改进增强算法。主要包括对彩色图像本身的增强、基于图像域的多曝光图像融合增强以及基于频率域的多传感器图像融合增强。本文的主要创新点：1.提出一种基于视觉感知的色调映射(Tone Mapping, TM)彩色图像增强方法。针对人类视觉的全局和局部感知自适应性，通过改进相应的S型色调映射函数，有效实现图像对比度增强。为克服直接在RGB彩色空间中做增强处理后色彩失真和不自然的感观效果，通过将原图像变换到La*b*颜色空间中再进行增强处理，实验结果表明该方法能保持彩色一致性和实现图像快速增强。2.提出一种结合Retinex理论和遗传算法(Genetic Algorithm, GA)优化的快速彩色图像增强方法。将基于灰度图像的传统图像增强方法直接推广应用到彩色图像增强，必然会破坏图像的彩色平衡，造成色彩感知的不协调性。为此，提出一种基于遗传算法优化的快速对比度增强算法，结合Retinex理论对图像的亮度进行处理，消除光晕，最后重建增强后的彩色图像克服了色彩不协调问题。实验结果表明使用所提出的方法增强后的彩色图像视觉质量得以提高且优于其他传统的技术方法，并展示了其有效性和鲁棒性。3.提出一种基于梯度信息的多曝光融合（Multi-exposure Fusion）高动态范围图像（High Dynamic Range Image, HDRI）合成方法。为克服通用多曝光融合增强中权值平均等方法中出现的不考虑信息重要性及邻域像素关系所造成的细节损失和模糊等问题，提出依赖于图像曝光质量评估及梯度域信息进行权值设计，对图像进行融合，实验结果表明增强的图像兼具原始图像在暗区和亮区的相应细节，图像整体效果符合人类视觉感知特性要求。4.提出一种基于分块去混淆（Ghost Removal）的多曝光融合算法。在处理动态多曝光图像融合增强过程中，最大的问题是如何解决由于运动所产生的混淆现象。本文通过提出基于梯度上升优化处理的自适应分块方法，并结合形态学原理，调整分块大小及动态区域块的权值，最终达到混淆去除的目的。同时利用Gaussian中心函数窗口滤波，去除在分块融合过程中引入的块边缘不连续性痕迹。实验结果表明该方法能有效增强多曝光图像并去除混淆问题。5.提出基于双变换的可见光视觉图像与红外图像增强方法。针对低光照或是伪装遮挡图像在可见光视觉图像与红外图像序列成像的相关特性及融合处理中存在的问题，提出一种基于亮度预增强处理的二代Curvelet与Harr小波双变换权值系数融合处理机制来有效保留边缘信息和图像细节信息的增强方法。实验结果表明所提出的方法提高了融合图像的视觉感知质量，增强后的图像为遮挡和伪装目标的检测和定位提供了更为有效空间环境。更多还原

【Abstract】 Image enhancement processing is related with a series of techniques to improveinformation of the focused region, weaken or remove the unnecessary information, oreven to convert the whole image into more suitable information model for human andmachine post-processing such as perception, analysis and recognition in the specialapplication fields. The design of the image enhancement algorithms is focused on thespecial goal of the application. The more requirements for higher quality images in suchimage applications have increased rapidly with the deeper development and widerapplications of the multimedia technologies and products. However, the visual effect ofthe image is affected by many internal and external conditions with the digital imagecapture device, dynamic range in the scenes and light condition. Sometimes thedegraded image is even hard to recognize for human and machine analysis. In order toobtain high quality images do with the processing of the image dynamic rangeadjustment, contrast stretch, color image enhancement and visual perceptionconsistency.Several theoretical methods and techniques of image enhancement are deeplystudied in the dissertation which based on the requirements of the special applications.And further improved algorithms are proposed in the dissertation by analysis of theirproblems and disadvantages in image processing applications. The main work of thedissertation includes enhancement of the color image itself, multi-exposure imagesfusion enhancement based on image domain, and multi-sense images fusionenhancement based on frequency domain.The main innovation of the dissertation:1. A true color image enhancement method is proposed based on the improvementof the visual perception with the tone mapping operation. The proposed methodimprove the contrast of the image adaptively by modified the S-shaped function to suitfor human eye percepts the image in global or local scene information. We firstlyconvert the original image from RGB color space into La*b*color space and do theenhancement in the luminance channel, and in the end combined with the chromaticinformation to overcome the color distortion and unnatural color perception effectsintroduced by the image processing directly in the RGB three channels. Colorconsistency maintaining and achieving adaptive fast image enhancement effects are provided in the resulting images by the proposed algorithm.2. A fast image enhancement method which combined Retinex theory and geneticalgorithm optimization is proposed for color image enhancement. The proposed methodcan disregard the shortcomings introduced by application the traditional imageenhancement methods on gray-scale image expand directly to color image enhancement.So use the Retinex method to decompose the brightness and refection information andapply the genetic algorithm optimization corresponding parameters to enhance theimage brightness information. Finally the proposed method compared with the othermethods in the enhanced result color images is more effective and robust for colorimage enhancement.3. A high dynamic range (HDR) imaging method is proposed by usingmulti-exposure images fusion on their gradient domain information. The proposedmethod takes advantage of the gradient indicates the significant information andneighbor pixels relationship among the multi images in order to design a better weightmatrix for improving visual quality of the fusion image, and overcome the loss ofdetails and fuzzy effects from the conventional weight-averaged methods, finallyacquires an enhanced image with details both improved in brighter and darker regions.4. A free of ghost multi-exposure fusion method is proposed based on image block.The key issue in the processing of multi-exposure fusion image enhancement indynamic scenes is how to remove the ghost due to the movement. The method based ongradient-ascent optimization perform appropriate block, and combine morphologicalopen/close operations, gradient magnitude and gradient direction relative differencebuild objective function for removing ghost. Then fusion image is filtered byGaussian-center blend function to smooth the block boundary artifacts.5. An image enhanced method based on dual-transform is proposed for visualimage and infrared (IR) image fusion. The method based on the secondgenerated-Curvelet and Harr wavelet transform acquired the fusion weight coefficient toeffectively preserve edge and details information in the images for visual and IR imageswith poor-light condition and targets camouflaged in the scenes. The enhanced resultsboth provide more clear space environment for target localization and better visualperception for target recognition.更多还原