Influence of image quality on SWT voting-based color reduction method for detecting text in natural scene images

The aim of the article is to study the influence of image quality on performance of text detection methods in natural scene images with emphasis on text segmentation stage. Natural scene images are often subject to numerous image-quality degradation factors, among which blur and noise are by far most common. To analyze their influence on performance of text detection methods, we systematically evaluated three state-of-the-art text detection methods, namely SWT, text detection-oriented color reduction and our own SWT voting-based color reduction, on the images that were degraded in the image capture process. Experimental results indicate that the SWT voting-based color reduction mostly outperforms the other two state-of-the-art methods. The experiment was carried out on a challenging CVL OCR DB text detection evaluation dataset.