基于Zernike矩的极光图像检测

Auroral Image Detection Based on Zernike Moments

摘要: 本文提出了一种基于zernike矩的极光图像检测算法。该算法首先对每幅图像提取zernike矩特征，然后用欧氏距离衡量图像之间的相似性，最后再依据相似程度用kNN方法进行分类。现实生活中的极光图像会不可避免的受到噪声因素的干扰，这就需要检测算法具有很好的鲁棒性。有时为了适应图像旋转的需求，也需要这种算法具有旋转不变性，而基于zernike矩的极光图像检测算法正好具备这些特点。与传统的LBP特征提取方法相比，由于连续正交的zernike矩是定义在极坐标下的，因此本身具有旋转不变性。通过实验表明，在图像添加噪声、旋转以及平滑滤波后，基于zernike矩的特征提取方法比LBP的检测效果好。

Abstract: In this paper, an auroral image detection algorithm based on zernike moments is proposed. Firstly, the zernike moments are extracted for each image, and then the similarity between the images is measured by Euclidean distance. Finally, the kNN method is used to classify them according to the similarity degree. In real life, the aurora image will inevitably be disturbed by noise factors, which requires the detection algorithm having a very good characteristic of robustness. Sometimes in order to adapt to the needs of image rotation, the algorithm is needed to have a characteristic of rotation invariance. The aurora image detection algorithm based on zernike moments just has these characteristics. Compared with the traditional LBP feature extraction method, since the continuous orthogonal zernike moments are defined under the polar coordinates, they themselves have the characteristic of rotational invariance. Experiments show that the feature extraction method based on zernike moments is better than LBP when the image is added with noise, rotation and smoothing.