Smooth Stroke Width Transform for Text Detection

• Published in: Artificial Intelligence: Methodology, Systems, and Applications pp. 183 - 191
• Main Authors: Oh, Il-Seok, Lee, Jin-Seon
• Format: Book Chapter
• Language: English
• Published: Cham Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Natural scene recognition; Regularization; Stroke width transform; Text detection
• ISBN: 3319447475; 9783319447476
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-319-44748-3_18

The stroke width transform (SWT) is a generic operation for the task of detecting texts from natural images because the characters intrinsically have the elongated shape of nearly uniform width. The edge pairing technique was recently developed by Epshtein et al. and is popularly used due to its simplicity and effectiveness. However since the natural images are noisy and sensitive to variations, high degree of artifacts arises and it hinders subsequent processing of the text detection. This paper reformulates the SWT problem in a new way that searches for an optimal solution in 3-D space. We present an effective search algorithm called the aggregation approach, borrowed from the depth image reconstruction domain. The experiments showed that the algorithm produced a smooth SWT map which is better for subsequent processes.