【作者】 周欣；

【导师】 林玉池；

【作者基本信息】 天津大学 ， 测试计量技术及仪器， 2009， 博士

【摘要】 红外热成像技术在军事和民用等很多领域发挥着越来越重要的作用,它拓展了人类视觉认知极限,将人类的视觉感知范围由传统的可见光谱扩展到视觉不可见的红外辐射光谱区。但是,红外图像的对比度不高,视觉效果模糊,不利于进行观察和提取景物特征信息。对红外图像进行增强处理,提高红外图像的对比度,改善图像视觉效果,成为目前红外热成像领域研究的一个重要方面。研究、分析和总结了图像增强方法,并针对目前图像增强算法中普遍存在的图像局部细节保护和噪声滤除之间的矛盾,提出了基于单幅图像多尺度方向分析的红外图像增强方法和基于多源图像融合的图像增强方法。多尺度方向分析是近年来在小波分析的基础上发展起来的图像稀疏表示方法。论文将多尺度方向分析理论应用于红外图像的去噪增强处理中,提出一种基于NSCT (Nonsubsampled Contourlet)变换的红外图像增强方法。利用NSCT变换在处理图像几何结构方面的优势,对红外图像的噪声和边缘信息分别进行处理,从而能够在增强图像边缘和细节信息的同时抑制图像噪声。NSCT变换不但继承了Contourlet变换的多尺度、多方向性,同时还具备了平移不变特性,因此能够更好地保持图像边缘。近年来,多源图像的使用导致了信息表现形式的多样性,将多源图像合成到一幅图像中显示,不仅适合图像进一步的处理和分析,而且更满足了人类视觉感受的需要。图像融合技术可以将在同一时间、或不同时间获取的关于某个具体场景的多源图像信息加以综合,并生成一个新的有关此场景的描述。红外和可见光成像传感器是两种功能原理不同的图像传感器。红外成像传感器检测目标发出的不可见热辐射,而可见光成像传感器则通过吸收目标反射的可见光波段电磁波来实现对目标的探测。通过将可见光图像和红外图像这两种不同类型图像的融合,得到了单幅图像增强技术所达不到的视觉增强效果。面向视觉增强的图像融合技术主要问题在于如何将多源图像中的视觉信息有效的合成到一幅图像中去显示,并同时抑制源图像中的噪声,以增强融合图像的视觉效果。将多尺度方向分析方法与多源图像融合技术结合起来,提出了基于统计特性的图像融合增强方法和基于区域特性的图像融合增强方法。考虑到传感器噪声的存在,设计了基于统计模型的多源图像融合增强方法,不仅更有效地融合了多源图像的信息,而且抑制了传感器噪声对图像融合的影响。人类视觉系统对图像清晰度的判断是由区域内像素共同体现的。根据NSCT变换后的低频子带和高频方向子带的特性,设计了基于区域特性的图像融合方法,在增强融合图像视觉效果的同时,有效地保持融合图像的细节和纹理信息。经过红外与可见光图像的融合实验表明,面向视觉增强的图像融合技术能够明显改善单一传感器的不足,提高图像信息的利用效率,从而更为准确和全面地获取对目标或场景的信息描述。基于单幅图像多尺度方向分析的红外图像增强方法和基于多源图像融合的图像增强方法各有自己的特点。前者是针对单一图像采用多尺度方向分解的方法进行的变换域图像处理,以去除图像中的噪声并增强细节和边缘等有用信息;而后者是将多个图像传感器采集到的图像进行融合以得到比单一图像传感器更丰富的图像信息。这两类图像增强方法适用于不同的图像处理环境,均能取得较好的增强效果。更多还原

【Abstract】 The technology of infrared thermal imaging plays a more and more important role in the military, civil field and other fields. It has enlarged the range of the human’s visual cognition from the visible spectrum zone to the invisible infrared spectrum region. However, the value of the contrast ratio of the infrared image is low, and the visual effects of it are not good. Therefore, the infrared image goes against observation and extraction of scenery’s feature. The enhancement of the infrared image for improving its contrast ratio and visual effects has become one important aspect of the current thermal imaging field.The dissertation has studied on the measures of image enhancement and summarized the current approaches. The dissertation has put forward the measure of image enhancement based on the multi-scale directional analysis of the individual image and the fusion of multi-source image. The measure aims to solve the contradiction between the protection of the image’s partial detail and the noise filtering.The multi-scale directional analysis is the sparse representation method based on wavelet analysis which has been developed in recent years. The dissertation has applied the multi-scale directional analysis to the de-noise and enhancement treatment of infrared image, and put forward the infrared image enhancement measure which is based on Nonsubsampled Contourlet transform. Taking the advantage of the Nonsubsampled Contourlet transform in the treatment of image’s geometrical structure, the infrared image’s noise and marginal information can be dealt with separately, therefore, the de-noise can be realized while the image’s marginal and detail information has been reserved. Nonsubsampled Contourlet transform has not only inherited the benefits of the multi-scale, multi-directional of the Contourlet transform, but also has the feature of translation invariant. It can reserve the image’s marginal information more effectively.In recent years, the application of multi-source image has resulted in the diversity of the image information manifestation. The integration of the multi-source image into one image can meet the requirement of human’s visual cognition and is beneficial to the further analysis by the computer. The image fusion technology can integrate the multi-source image about certain concert scenery acquired at the same time or at different time into a new description about the scenery. Infrared and visible light imaging sensor are two different kinds of image sensors with different function and principle. The infrared imaging sensor can detect the invisible thermal radiation, while the visible light sensor can detect the target through the absorbing of the visible light band electromagnetic wave reflected by the target. Through the fusion of the visible light image and the infrared image, we can achieve much better visual effects than those of the individual image enhancement technology.The problem of the image fusion technology merely working in the visual enhancement rests with how to effectively synthesis the visual information in the multi-source image into one image and how to de-noise the source image to enhance the fused image’s visual effects. The dissertation combines the multi-scale directional analysis and the multi-source image fusion approaches, and brings forward the image fusion enhancement measures based on the statistical characteristics and the regional features respectively.Considering the existence of sensor’s noises, the multi-source image fusion enhancement measure which is based on the statistical model has been put forward. The measure not only more effectively fuses the information of the multi-source image, but also restrains the influence of sensor’s noises to the fused image.The judgment of the human’s visual system towards the image definition is commonly embodied by the pixels in the region. According to the characteristic of low frequency sub-band and high frequency directional sub-band after the NSCT transform, we designed the image fusion measures based on regional features, which enables the effective reservation of the marginal and detail information while enhancing the fusion image visual effects.According to the infrared and visible image fusion experiments, the image fusion technology aiming at improving the image visual effects can make up for the deficiency of single sensor, improve the use efficiency of image information, therefore, provide a more accurate and comprehensive description towards the target or the scene.The infrared image enhancement measure based on the individual image multi-scale directional analysis and the image enhancement measure based on the multi-source image fusion have their own advantages and disadvantages. The former is aimed at dealing with individual image which has adopted the multi-scale directional analysis in carrying out the frequency domain treatment to de-noise and enhance the marginal and detail information; the later is to fuse the image collected by the multi-source image sensors to acquire the image information which obtains more information than the individual image sensor. The two image enhancement measures apply in different image treatment environment, and both can acquire better effects of enhancement.更多还原