Adaptive Scale Estimation And Robust Entropy-like Model Fitting Algorithm

Robust model fitting techniques have been widely used in many computer vision problems, such as line and circle fitting, homography matrix and fundamental matrix estimation, motion segmentation, and range image segmentation, etc. In practice, data are always contaminated (duo to the factors such as faulty feature extraction, sensor noise, segmentation errors, etc.), model fitting is a non-trivial task. The accuracy of the scale estimation will affect the results of model fitting and segmentation. However, the-state-of-the-art scale estimators may be invalid when outlier percentage is very large. In this thesis, to improve the robustness of model fitting, we propose a robust scale estimator called AIKOSE and two robust model fitting algorithms called AMSAC and ASEE.The innovative contributions are two-fold:In order to solve the problem of adaptive scale estimation, we propose a novel robust scale estimator called AIKOSE. It can estimate the scale of inlier noise by adaptively selecting the optimal value of K in the IKOSE scale estimator. Experimental results show that AIKOSE is very robust in scale estimation.In order to solve the problem of parametric model estimation, we propose a robust estimation method called ASEE (the Adaptive Scale based Entropy-like Estimator), which minimizes the entropy of inliers. This estimator is based on IKOSE and LEL (the Least Entropy-Like estimator). Unlike LEL, ASEE only considers inliers’ entropy and excludes the outliers, which makes it very robust in parametric model estimation and can deal with the data containing up to90percent outliers. Compared with other robust estimators, ASEE is simple and computationally efficient. Experimental results on both the synthetic and real-image data show that ASEE is more robust than several state-of-the-art robust estimators, especially in handling extreme outliers. Moreover, based on AIKOSE, we propose a novel robust estimator called AMSAC, which can fit a model without requiring a manually tuned K value. In the experiments, we demonstrate the robustness of AMSAC on line fitting and homography estimation by using both synthetic data and real images. The results show that AMSAC is more robust than other competing robust estimators and can handle the data involving up to90percent outliers.