Text detection in scene images using stroke width and nearest-neighbor constraints

Text in scene images can provide very useful as well as vital information and hence, its detection and recognition is an important task. We propose an adaptive edge-based connected-component method for text-detection in natural scene images. The approach is based on three reasonable assumptions - (i) characters of a particular word are locally aligned in a certain direction (ii) each character is of uniform color ( iii) stroke width is almost constant for most of the characters in a particular word. We apply color quantization and use the luminance to obtain the intensity values. An improved edge-detection technique that performs adaptive thresholding is used to capture all possible text components with some non-text components initially. Then, we remove obvious non-text component based on a few heuristics. Further, we classify those components as text for which we successfully obtain two consecutive nearest-neighbors that are aligned in a direction and satisfy certain constraints based on size and inter component distance. Finally, we estimate stroke width and foreground color for each component and those having a fairly uniform value of the same are classified as text. Results on ICDAR 2003 Robust Reading Competition data show that the method is competitive for text-detection. The main advantage of our method is that it is robust to font-size, degraded intensities and complex backgrounds. Also, the use of stroke width and color in this manner for text detection is novel to the best of the authorpsilas knowledge.