遥感图像光照一致性校正

发布时间：2018-08-07 22:10

【摘要】：光照条件是图像成像过程中的重要条件之一,影响图像的质量,很多情况下要求图像具有光照一致性,如高低分辨率图像对比法计算图像的MTF值、遥感图像拼接、人脸识别、颜色恒常性研究等领域,因此图像的光照一致性问题引起国内外学者的广泛研究。本文的研究目的是提出一种有效的多幅图像光照一致性校正方法,能够使不满足光照一致性的图像经过处理后,满足这一条件,从而应用于各领域。本文的研究思路是首先通过单幅图像光照校正方法消除单幅图像内的光照不均,然后通过多幅图像间的光照校正方法使图像间光照一致,最后再通过定量评价方法结合主观视觉感受来评价多幅图像的光照一致性和校正方法的有效性。通过观察大量的真实遥感图像,发现图像的光照情况可分为几大类,偏暗、偏亮或对比度低等情况,若直接对这些图像进行光照一致性校正,效果并不理性,因此要先通过单幅图像的校正方法消除图像内的光照不均。通过实验比较各种单幅图像校正方法的效果,最终选择梯度域方法作为我们的单幅图像校正方法,该方法适合于各种光照情况的校正,且效果均比较理想。多幅图像间的光照一致性校正方法并没有现有的方法可供选择,因此我们提出了自适应的伽玛校正方法和改进的奇异值均衡化方法,通过将图像间的亮度调整一致,使得图像的动态范围最大程度的重合,从而增强图像间的光照一致性。本文也首次提出了一种多幅图像光照一致性定量评价方法,该方法首先提取光照图像,然后比较光照图像直方图的相似度,将这一相似度约束在0到1之间,1表示一致性最好的情况。实验结果表明,本文提出的多幅图像光照一致性校正方法能够改善图像间的光照一致性,主观评价和客观定量评价均能取得较好的结果。本文还给出了大量实验的统计数据,用统计数据更有力地说明该方法的有效性。
[Abstract]:Illumination condition is one of the important conditions in the image imaging process, which affects the quality of the image. In many cases, it requires the image to have illumination consistency, such as the high and low resolution image contrast method to calculate the image MTF value, remote sensing image mosaic, face recognition. Color constancy is studied in many fields, so the illumination consistency of image has been studied widely at home and abroad. The purpose of this paper is to propose an effective illumination consistency correction method for multiple images, which can make the image which does not satisfy the illumination consistency satisfy this condition after processing, so it can be applied in various fields. The idea of this paper is first to eliminate the uneven illumination in a single image by using a single image illumination correction method, and then to make the illumination between images consistent by the illumination correction method between multiple images. Finally, quantitative evaluation method combined with subjective visual perception is used to evaluate the illumination consistency of multiple images and the effectiveness of the correction method. By observing a large number of real remote sensing images, it is found that the illumination of the images can be divided into several categories, such as dark, bright or low contrast, etc. If the illumination consistency of these images is corrected directly, the effect is irrational. Therefore, the correction method of single image should be used to eliminate the uneven illumination in the image. Finally, gradient domain method is chosen as our single image correction method. The method is suitable for various illumination conditions, and the effect is satisfactory. There are no existing methods for illumination consistency correction among multiple images, so we propose an adaptive gamma correction method and an improved singular value equalization method. The dynamic range of the image is overlapped to the maximum extent, which enhances the illumination consistency between the images. In this paper, a quantitative evaluation method of illumination consistency is proposed for the first time. Firstly, the illumination image is extracted, and then the similarity of the histogram of illumination image is compared. Constraining this similarity between 0 and 1 indicates the best consistency. The experimental results show that the illumination consistency correction method proposed in this paper can improve the illumination consistency among images. Both subjective evaluation and objective quantitative evaluation can achieve good results. In this paper, a large number of experimental statistical data are also given to illustrate the effectiveness of the method.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【参考文献】