A new stroke filter by DOG
Chu‐Sing Yang, Yung-Hsian Yang · 2016
We propose a region base simple method to filter out strokes, which is brighter or darker than the background. And this method had good performance at blur, low contrast, and shadow. The method is implement by Difference of Gaussian (DOG), which is applied in edge detection, invariant feature. We use the properties of DOG, isotropic edge detect and sign value than background, to develop a stroke over scale. Then we use a dynamic threshold method and a static threshold value to filter out the noise region. After morphology process, we can filter out text as regions. In result part we test our method by DIBCO, ICDAR, and license plate in real scene to prove our method have more robust in harsh environment.