A Modified RANSAC mechanism: Multiple models extraction algorithm

摘要

As one of the most frequently used regression methods, RANSAC is advanced in its effectiveness and efficiency, but RANSAC cannot extract multiple models due to its exclusivity. To extract multiple models, this paper proposes a new regression method, which is a modified version of RANSAC. To fit a model to data points, a labeling process classifies each data point into proper inlier, quasi-inlier or outlier. The model is obtained from the proper and quasi inliers. After eliminating the proper inliers, another model fitting is performed. These operations are repeated till no more model is fitted. The effectiveness of the proposed method is shown by experiments on extracting multiple lines from images.