基于相位一致性的太阳图像低对比度特征提取研究

There are abundant low-contrast phenomena in solar observations. Traditional method of feature detection using intensity or/and gradient threshold is extremely diffcult to accurately identify the low-contrast features, because they appear as blurred edges and relatively low intensity values. Phase Congruency (PC) is a novel method describing low-level features of image, which is perfectly combined local image intensity with phase in frequency domain. It is invariant measure to image luminance, contrast, intensity, and image magnification. The characteristics can solve the problem that traditional method of image processing excessively depends on image intensity or/and gradient threshold.Our expriments show that the feature detection with PC is more efficient than traditional methods with image intensity and gradient in respect of low-contrast feature extraction. We will research the low-contrast features with sunspot umbra and conoral loops of off-limb because they exhibit typical features of lower contrast than adjacent features.We focus on solving challenging technical problems in obtaining phase characteristics, noise compensation and description of ROI using PC. One of the innovative points of the proposal is that we lead the PC technique to sloar image processing in order to solve the problem of low-contrst feature detection. The goal of this proposal is to increase the identification accuracy of low-contrast features and the robustness of the algorithm from a new perspective, and to reduce the negative effects of subjective factors in solar image processing. Finally, the proposed research may be applied to other features extraction from solar images.

太阳观测图像中存在大量的低对比度特征结构。它们具有局部相对强度小、结构边缘模糊的特点，因而难以运用传统的基于强度和梯度两个基本量的图像处理方法进行准确的特征结构提取。近年来的研究发现，相位一致性也是一种描述图像底层特征的基本量，并且对特征强度和对比度的测度具有不变性。我们的实验表明，与传统的识别方法相比，基于这一理论的处理方法在太阳图像低对比度特征结构提取上有较为明显的效果。本项目以太阳黑子本影区域的本影点和本影闪耀，以及冕环的形态特征作为具体的研究对象，进一步深入研究如何将相位一致性理论和方法应用于多种太阳观测图像，以便有效地提取低对比度特征结构。本项目拟解决的关键技术包括：滤波器的设计以及方向的选择、噪声补偿机制、特征提取与所测量物理量的联系等。项目特色和创新在于将与强度和梯度图像处理方法完全不同的底层特征描述- - 相位一致性引入到太阳图像处理中。本项目研究成果预计将有广泛的应用价值。

目前大量的高时间高空间太阳观测数据，揭示了大量的低对比度太阳特征结构，如本影点、冕环、色球纤维和针状体等。从视觉上，它们具有局部相对强度小、结构边缘模糊的特点，难以运用传统的基于强度和梯度的图像处理方法来进行准确的特征结构提取。相位一致性是一种描述图像底层特征的基本量，对特征强度和对比度的测度具有不变性。项目以太阳黑子本影区域的本影点（UDs）和光球米粒，以及冕环的形态特征作为具体的研究对象，进一步深入研究如何将相位一致性理论和方法应用于多种太阳观测图像，有效地提取低对比度特征结构，发展具有自主知识产权的太阳观测数据处理方法。项目重点研究了（1）太阳黑子本影点的特征提取和跟踪算法，分析了它们的统计特征和振荡行为；（2）太阳光球米粒的的特征和跟踪算法，以及它们的形态特征演化以及生命期；（3）冕环特征提取算法以及振荡行为特征分析等低对比度特征提取算法。项目以NVST和SDO观测数据为基础展开，解决了传统方图像处理技术在提取低对比度特征结构时所面临提取难、跟踪难等问题。同时还对太阳观测图像的配准算法，太阳观测图像的噪声抑制算法，以及准周期信号的时频分析技术进行了初步的探索，并取得了新的科学结果。在项目执行期间共发表了12篇SCI和EI期刊论文，获得了4项具有自主知识产权的太阳图像处理方面的技术发明专利。