【摘要】：近年,圖像配準技術廣泛應用于建筑學、醫(yī)學以及軍事學等領域。現(xiàn)實應用中,由于獲取所需圖像的整體面貌比較困難,所以可以先獲取局部圖像再運用圖像配準技術將多幅局部圖像拼接成所需圖像的整體面貌。圖像配準的速度與準確率是目前圖像處理領域中的熱點與難點。如何提升圖像配準的速度和圖像拼接的質(zhì)量在各研究領域中有著重要的意義。圖像配準即針對兩幅或多幅有共同部分的圖像,對它們進行匹配與拼接來得到一張完整圖像的技術。實現(xiàn)該技術的方法大致分為三類,分別基于圖像的灰度信息、變換域以及圖像的特征。本文中的SIFT(Scale-invariant feature transform)算法即是基于圖像中的特征進行圖像配準的算法。該算法在圖像的不變特征提取方面效果很好,但對于某些特定的圖像,如:噪聲多、像素高及某些信息位置集中的圖像,該算法的運算速度會較慢,不能很好的保證高實時性的特定要求,仍然存在不足。本文首先對SIFT算法及其變種算法的原理與過程進行了深入討論,并進行實驗過程對比分析,在此基礎上,針對SIFT算法應用于某些特定圖像運算速度上的不足進行改進。即運用SVD(Singular value decomposition)方法對初始圖像進行壓縮處理,通過壓縮,選取合適的特征值數(shù)量,使得圖像中提取的特征點描述子減少,從而減小算法的運算量,達到減少運算時間的目的。通過數(shù)值實驗結果分析,驗證了圖像壓縮處理過程中使用特征值的數(shù)量與算法提取到的特征點描述子的總數(shù)量成正比關系,以及特征點描述子的總數(shù)量與算法的平均運算時間成正比關系,即本文的改進提升了SIFT算法的運行速度。改進后的算法在保證原有準確率的同時,計算速度比原有算法更快,為圖像配準的工作提供了很好的保障。
[Abstract]:In recent years, image registration technology has been widely used in architecture, medicine, military and other fields. In practical application, because it is difficult to obtain the overall appearance of the required image, it is possible to obtain the local image first and then use the image registration technology to mosaic multiple local images into the overall appearance of the required image. The speed and accuracy of image registration is a hot and difficult point in the field of image processing. How to improve the speed of image registration and the quality of image mosaic is of great significance in various research fields. Image registration is the technique of matching and stitching two or more images with common parts to get a complete image. The methods of implementing this technology are divided into three categories, which are based on the gray information of the image, the transform domain and the characteristics of the image. The SIFT (Scale-invariant feature transform) algorithm in this paper is an image registration algorithm based on the features in the image. The algorithm has a good effect on image invariant feature extraction, but for some specific images, such as high noise, high pixel and some information position sets, the operation speed of the algorithm will be slow. There are still shortcomings in the specific requirements that can not guarantee high real-time performance. In this paper, the principle and process of SIFT algorithm and its variant algorithm are discussed in detail, and the experimental process is compared and analyzed. On this basis, the application of SIFT algorithm to some specific image operation speed is improved. That is, the SVD (Singular value decomposition) method is used to compress the initial image, and the appropriate number of eigenvalues is selected to reduce the number of feature points described in the image, thus reducing the computational complexity of the algorithm. To achieve the purpose of reducing operation time. Through the analysis of numerical experimental results, it is verified that the number of eigenvalues used in image compression processing is proportional to the total number of feature point descriptors extracted by the algorithm. And the total number of feature point descriptors is proportional to the average operation time of the algorithm, that is to say, the improvement of this paper improves the running speed of SIFT algorithm. The improved algorithm not only ensures the original accuracy, but also calculates faster than the original algorithm, which provides a good guarantee for image registration.
【學位授予單位】：吉林大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP391.41

【相似文獻】

相關期刊論文 前10條

1 王艷陽;傅小龍;夏冰;吳正琴;樊e，

本文編號：2486959