Gabor filtering-based scale and rotation invariance feature for 2d barcode region detection

2D barcode region detection is a non trivial problem for barcode revognition and decoding especially in complex backgrounds. Currently, morphological processing has been widely applied to extract potential regions of Data Matrix barcodes due to its low computation complexity. However, this method leads to two problems, adaptive selection of morphological structuring element and high false accept rate. To solve these problems, this paper proposes an innovative method for 2D barcode region detection based on Gabor filtering and BP neural network. The contributions are two folds: 1) we propose a texture feature formulation independent of scale and rotation; 2) BP neural network can avoid the difficulty in morphological structure construction. Large scale experiments show the accuracy and robustness of the proposed method over the traditional morphological method.