基于Retinex方法的无人机影像阴影去除应用研究

发布时间：2018-12-29 15:36

【摘要】：由于光照、地形起伏等原因,无人机航测获取的影像中会出现阴影,使得影像对比度降低、地表信息模糊,降低了影像质量同时也降低了 DOM质量和目视解译的准确度。为了有效恢复无人机航测影像中阴影遮蔽下的重要地表信息,论文分析国内外关于航空影像阴影消除研究进展的基础上,以陇南山区土地确权项目中获取的存在阴影的无人机航测影像为数据源,选取10幅影像作为代表数据并基于实验区影像中阴影的性质和特点,利用传统处理方法和Retinex方法对存在阴影的影像进行处理并进行对比分析研究,具体的研究结论如下:(1)利用代数运算、直方图均衡化、同态滤波、匀光匀色等传统图像增强方法对存在阴影的影像进行处理,结合主观效果以及标准差、信息熵、平均梯度影像质量指标统计值对实验结果进行综合对比评价。结果表明:相对于原始影像,利用传统方法处理影像会使影像质量指标值有所增加,阴影区域地表信息有所恢复。(2)利用基于RGB色彩空间下的Retinex算法对存在阴影的影像进行处理,具体实现单尺度(SSR)、多尺度(MSR)和带颜色恢复因子的Retinex算法(MSRCR)并进行实验,通过对高斯模糊尺度参数的不同取值的处理结果进行对比分析,探究出相对合理的取值。结果表明:将MSRCR算法应用于无人机影像的阴影去除,虽然处理后的部分影像有一定的色彩偏离,但是在处理效果明显性、细节恢复程度等方面的效果较为显著,同时较原始影像,统计其影像质量指标平均值,平均梯度、标准差、信息熵值分别提高了 4.5倍、1.5倍、0.2倍。(3)针对RGB色彩空间下的Retinex算法应用中出现色彩偏离的问题,利用基于HSV色彩空间下的Retinex算法对存在阴影的影像进行处理,并改进算法中的像素值量化方式,最后通过特征匹配实验验证算法可行性。结合主观效果和影像质量评价指标进行综合评价,结果表明:一定程度的解决了色彩偏离的问题,有效恢复了阴影遮蔽下的地表信息,同时较原始影像,统计其影像质量指标平均值,平均梯度、标准差、信息熵值分别提高了 16倍、3.4倍、0.3倍。通过上述的研究,表明利用Retinex方法可以有效恢复阴影遮蔽下的地表信息,该方法应用于无人机影像的阴影去除是可行的。
[Abstract]:Because of illumination, terrain fluctuation and so on, shadow will appear in the images obtained by UAV aerial survey, which makes image contrast lower, surface information blurred, image quality reduced and DOM quality and visual interpretation accuracy reduced. In order to effectively restore the important surface information of aerial vehicle aerial survey image under shadow shade, this paper analyzes the research progress of shadow elimination of aerial image at home and abroad. Taking the aerial vehicle images with shadows obtained from the land certification project in Longnan mountainous area as the data source, 10 images are selected as the representative data and based on the nature and characteristics of the shadows in the images of the experimental area. The traditional processing method and Retinex method are used to process the shadow images and the results are as follows: (1) using algebraic operation, histogram equalization, homomorphic filtering, Traditional image enhancement methods such as uniform light and uniform color are used to process shaded images. The experimental results are compared and evaluated comprehensively with subjective effects, standard deviation, information entropy and statistical value of average gradient image quality index. The results show that compared with the original image, the image quality index will be increased by using the traditional method to process the image. (2) using Retinex algorithm based on RGB color space to process shadow images, and realizing single-scale (SSR),. (2) the surface information of shadow region is recovered. (2) the Retinex algorithm based on RGB color space is used to process shadow images. Multi-scale (MSR) and Retinex algorithm (MSRCR) with color recovery factor are experimented. By comparing and analyzing the processing results of Gao Si's fuzzy scale parameters, a relatively reasonable value is found out. The results show that the MSRCR algorithm is applied to the shadow removal of UAV images, although some of the processed images have color deviation, but the processing effect is obvious, the degree of detail recovery and so on. At the same time, compared with the original image, the average image quality index, the average gradient, the standard deviation and the information entropy were increased by 4.5 times, 1.5 times, respectively. (3) aiming at the problem of color deviation in the application of Retinex algorithm in RGB color space, the Retinex algorithm based on HSV color space is used to process the shaded image, and the pixel quantization method in the algorithm is improved. Finally, the feasibility of the algorithm is verified by feature matching experiment. Combined with subjective effect and image quality evaluation index, the results show that: the problem of color deviation is solved to a certain extent, the surface information under shadow is recovered effectively, and the original image is more effective than the original image. The average image quality index, average gradient, standard deviation and information entropy were increased by 16 times, 3.4 times and 0.3 times respectively. Through the above research, it is shown that the Retinex method can effectively restore the surface information under shadow shading, and it is feasible to apply this method to the shadow removal of UAV images.
【学位授予单位】：西安科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】