【作者】 谯帅；

【导师】 孙韶媛；

【作者基本信息】 东华大学 ， 控制理论与控制工程， 2012， 硕士

【摘要】 由于夜视图像为对比度较低的单色图像,灰度等级有限,彩色化的夜视图像可通过人眼的色彩感知获取更丰富的场景信息,从而改善场景的深度感知,提高目标识别和探测效率,减少视觉疲劳和判断时间。因此彩色夜视技术成为当前国内外夜视领域的研究热点,在军用及民用方面都有广泛应用。和静止的夜视图像相比,动态的夜视视频序列中隐藏着运动信息,将夜视视频序列彩色化可以更有效的将夜视技术服务于军用及民用。本课题在静态夜视图像彩色化的基础上,研究动态夜视视频序列的彩色化技术,从而扩大彩色夜视的实用性。论文首先对夜视图像的成像原理给予了阐述,并介绍了夜视图像不同于普通灰度图像的特征。对夜视图像的经典彩色化算法进行了总结和探讨。其次,针对同时获取的同一场景的红外(>700nm)和微光(<700nm)图像,提出基于颜色查找表的双波段视频彩色化算法。首先由微光与红外图像融合后通过色彩传递的方法,得到具有突出目标和清晰背景细节的自然感彩色图像,建立初步的颜色查找表,然后采用色彩修补的方式得到完整的颜色查找表。此后的双波段视频序列,在查找表中找到对应的色彩值,实现快速彩色化。由于使用颜色查找表,省去了空间转换和匹配传递的过程,计算量大大减少。并且颜色表中的颜色值是由色彩传递得到的,彩色化后的图像具有较真实的自然感。最后,针对车载红外视频序列,提出一种快速彩色化的方法,利用轮廓特征点跟踪来获取每帧物体类别的轮廓区域,采用类别特征色彩对各区域传递色彩。首先构建各景物样本特征色彩集,以各类景物在自然彩色图像中表现出来的特征色彩,作为夜视图像中对应景物的色彩；利用改进的高效K-Means方法对红外关键帧进行聚类,得到较好的分割区域,提取轮廓特征点；通过KLT算法跟踪特征点,得到其在下一帧中的位置并及时修正,采用B样条插值进行轮廓复原,得到该帧的各类别轮廓区域。对每帧区分好的类别区域,将特征色彩按类别赋予该区域,从而给物体着上合适的颜色,实现红外视频序列的快速彩色化。实验结果表明,该方法提高了处理的速度,能够得到较理想的效果。更多还原

【Abstract】 Natural sense of color night vision can be perceived by the human eye for richer scene information to improve the depth perception of the scene, to improve target identification and detection efficiency, to reduce visual fatigue and to determine the time. Therefore, color night vision technology has become the night vision research focus in the military and civilian aspects that widely used. Compared to static night vision image, dynamic night vision video sequence hide motion information, colorization of the night vision video sequence can be more effective serve military and civilian. This issue mainly research color night vision video sequences of dynamic technology on the basis of color of night vision image, thus expanding the usefulness of color night vision.Firstly, this paper gives a detailed exposition on night vision imaging process, and thus a detailed explanation of night-vision image which different from the normal gray-scale image features. The classic colorization algorithms of night vision image are summarized and discussed.Secondly, while access to the same scene for the infrared(>700nm) and low light (<700nm) images is proposed based on the color lookup table algorithm dual-band color video. First, by the low light and infrared image fusion method, after passing through the color to get outstanding goals and a clear sense of the background details of the natural color images, the establishment of the initial color lookup table, and then use color fix way to get the full color lookup table. After the dual-band video sequence, in the lookup table to find the corresponding color value, fast color-based. Then use of color lookup tables, eliminating the need for space conversion and matching transfer process, the calculation is greatly reduced. And the color values in the color table are obtained by the color transfer, color of the image with a more realistic natural sense.Finally, we proposed a fast method to colorize the vehicle infrared video, using contour feature points to track objects in each frame to get the outline of the regional categories with category colors to transfer color for each region. First, we build the color set of each sample scene, to various types of scenery in a natural color image color characteristics shown, which corresponding to the scene as night vision color; then use of improved and efficient K-Means clustering method to segment infrared key frame, get a better partition and extract the contour feature points; by KLT feature point tracking algorithms to get their positions in the next frame and timely corrected, use B-spline interpolation to recover contours, get the frame outline of each category area. Lastly, colorize each characteristic region by category colors. Experimental results show that this method improves the processing speed and can get more satisfactory effect.更多还原