Efficient Barcode Localization Method for Low-Quality Images

Barcode technology is widely applied to industrial automatic identification field for its low cost and high reliability as well as the fast real-time performance. Due to the complexity of low-light, rotation and blur in the industrial field, several barcode localization approaches which are superior in speed or accuracy fail to accurately locate and even detect the barcode. This paper proposes a real-time barcode approach that can effectively cope with the above problems when dealing with low-quality images. First, we rely on the gradient information of the pixels to obtain both orientation map and magnitude map. Then, we use the Shannon entropy theorem to get a salient map for the sake of segmenting salient patches of the high score. Later, we utilize the smoothing filter to remove the noise and connect the salient patches to form the barcode candidate BLOBs (Binary Large OBject). Finally, the correct barcode is selected from the above candidate BLOBs with a covariance matrix. We obtained 500 experimental images covering the conditions of reflection, rotation, and low illumination from the industrial site. The experimental results based on the dataset show that our method exceeds significantly the other three advanced methods in accuracy.