Dissertation
Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device

Research on Image Registration Method with Stabilization and Affine Invariance

Author MaLiTao
Tutor YangDan
School Chongqing University
Course Applied Mathematics
Keywords Image registration Condition number RANSAC Convex hull
CLC TP391.41
Type Master's thesis
Year 2008
Downloads 170
Quotes 1
Download Dissertation

Image registration is a basic problem in the field of image processing, it will obtain the same scene at different times, different sensors or under different conditions, two or more images match, the superposition process. Image registration is a variety of image processing and applications such as target identification, change detection, the basis of three-dimensional modeling, due to its wide application in many areas and complexity, it has become a hotspot of digital image processing One of the problems. Feature-based image registration method is one of the most common methods of image registration. It has two important aspects: First, the edge extraction of feature points, the second is a measure of the similarity that feature matching criteria. Both feature-based image registration key. Therefore, this paper on these two aspects with a focus on improvement. Important as the image of a localized feature extraction stage, corner points in the feature point, while retaining the important feature of the object information in an image while effectively reducing the amount of data of the information to become an important image feature in the image understanding and pattern recognition. We will study focused on how to filter out a more stable and more efficient feature points on the basis of existing corner, we introduce the concept of condition number, it is the stability of the corner of a quantitative analysis, delete unstable corner points, which can effectively overcome the effect of errors due to noise and other factors, thus improving the performance of the entire matching algorithm. Matching the selection criteria stage, the existence of the error matching points on the commonly used method of matching transform estimates difficult or even fail. Therefore, we have chosen the RANSAC algorithm has a good performance in this case considering mis-existing well to eliminate the impact, and recognize them, and then based on the transformation evaluation function to calculate the optimal transform. The same time, according to the RANSAC algorithm thinking, focus on the convex hull of the unique affine invariant local controllability proposed the RANSAC algorithm imitation of the convex hull, greatly reducing the computational burden and improve the computing speed, and the convex hull the formative stages of a fast convex hull algorithm based on the slope to make the registration process more efficient. Finally, through a large number of experiments and analysis of the proposed image registration algorithm based on the condition number of the convex hull, effectively reducing image registration algorithm based on the condition number of false match rate, improve registration accuracy, and based on convex hull image registration algorithm has good performance in reducing the time of registration, under the premise of ensuring registration accuracy, greatly improved the speed of the registration, and have reached the experimental results.

Related Dissertations
More Dissertations