Evaluation of Gradient Descent Optimization method for SAR Images Co-registration

2020 12th International Conference on Electrical Engineering (ICEENG)(2020)

引用 3|浏览1
暂无评分
摘要
Registration or matching process aims to find the misalignment between two or more images concerning the same area to detect the values of the mapping matrix in order to transform interest points in one image to its correspondence in the others. This paper presents a dynamic approach aiming to improve the performance of the registration process for synthetic aperture radar (SAR) images. First, the noise resulting from the capturing process is reduced by using a smoothing filter based on kernel-gaussian to reduce the amplification of noise. Then; a combination of two area- based matching (ABM) methods is used. The first method is carried out using Crosscorrelation approach, acting as coarse registration step. The second method is achieved by using regular step gradient descent (RSGD) optimizer, acting as fine registration step. Evaluation of the performance concerning the proposed manner is achieved by comparing to the state-of-the art detectors as Harris, Shi-Tomasi, and Features from Accelerated Segment Test (FAST) detectors. Metric factors to achieve the comparison are mean square error (MSE) and peak signal-to-noise ratio (PSNR) between the input images. Results demonstrate a highly performance for the proposed method compared to the others where it has a high robustness and minimizes the noise of the input image.
更多
查看译文
关键词
SAR images,image matching,feature detector,area-based matching,feature-based matching,transformation matrix,random sample consensus (RANSAC) algorithm
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要