Outlier detection with MSTOF for dot matrix character location

Abstract

For distinguishing outliers from targets and locating dot matrix character, outlier detection with mean shift trail outlier factor (MSTOF) is proposed to indicate the score of outlier-ness. Firstly, k-distance neighborhood of an object is employed and k-mean shift trail vector of an object is established in terms of the difference between the near two k-mean shift vectors. Secondly, k-average mean shift trail distance of an object is presented on the basis of the weighted sum of k-mean shift trail distances sorted in descending order from 1 to k. Finally, MSTOF response value of an object is calculated using its k-average mean shift trail distance and its corresponding k-distance neighbors. Experimental results demonstrate that the proposed algorithm can effectively identify local outliers and locate dot matrix character with high quality.