Abstract: A robust fundamental matrix estimation algorithm based on kernel fuzzy clustering is proposed. The residual features of match points are detected by this algorithm. Using the kernel functions, the nonlinear dividable features in the original one-dimensional space can be mapped to a high-dimensional feature space, in which the clustering can be performed effectively. The match points are classified into inliers and outliers by the fuzzy clustering method. The inliers and the outliers are modeled by Gaussian function, and the judgment values of the divisibility of the two sets are defined and calculated. The iteration will continue until the judgment value converges. Experimental results of the fundamental matrix computation on both simulated and real data demonstrate the superiority of the proposed algorithm in precision and efficiency over classical RANdom SAmple Consensus (RANSAC) algorithm.

Key words: computer application, fundamental matrix, robust, kernel fuzzy clustering, feature match