A Comparative Study of Feature Selection for SVM in Video Text Detection

Abstract

In this paper, a comparative study with three support vector machines (SVM) classifiers was carried out. The input images were first preprocessed to form the candidate text string regions. Next, Based on different features sets extracted by different methods, three SVM classifiers are used to analyze the textural properties of text and classify the text and no text strings in video frames. Then, a comparative evaluation of their performance is presented. The goal of the paper is to identify good feature selection for SVM in video text detecting task.