Detecting and Segmenting Text from Natural Scenes with 2-Stage Classification

This paper proposes a novel learning-based approach for detecting and segmenting text from scene images. First, the input image is decomposed into a list of connected-components (CCs) by color clustering algorithm. Then all the CCs including text CCs and non-text CCs are verified by a 2-stage classification module, where most of non-text CCs are discarded by cascade classifier and the remaining CCs are further verified by SVM. All the accepted CCs are output to generate result image. Experiments have been taken on a lot of images with different nature scenes and show satisfactory performance of our proposed method