A new method for displayed mathematical expression detection based on FFT and SVM

摘要

Extracting and recognizing mathematical expressions of scientific documents are key steps in the process of mathematical retrieval system, where the documents contain different components such as text, tables, figures, and mathematical expressions. There are several methods proposed to handle the components of documents. Those methods have investigated the feature of components based on the segmented text lines. This paper proposes a new method to detect mathematical presentation in the form of displayed expression from scientific documents using Fast Fourier Transformation. The test for the proposed method shows that the accuracy is from 94% to 98% for recognizing mathematical expressions on the benchmark datasets. It demonstrates the effectiveness of the proposed method in compared with the other ones.