Stroke detection and reconstruction of characters pressed on metal label

In order to detect features of protuberant characters, a novel stroke detection method based on Gabor filters is proposed. First, the gray images of protuberant characters were preprocessed using morphological algorithm. Next, a set of Gabor filters is used to break down an image of protuberant characters into four directional images, which contain the stroke information of four directions. Then, a reconstruction experiment is carried out with the Gabor characters. The results show that the Gabor representation has strong reconstruction power. Finally, A BP neural network is introduced to classify the Gabor features and the experiment results tell that the Gabor features have good separate capability. All of the above proves that the proposed method can be reliably used for feature extraction of pressed characters in low-quality images.