Two-dimensional Barcode Image Processing Based On Wavelet Decomposition And Neural Network

Information can be expressed through2D Barcode from both horizontal andlongitudinal dimensions. Featured by huge capacity of data storage, high informationintensity, high reliability, powerful error correcting capability and multi-informationexpression, QR Code plays an increasingly important role in automatic identification.QR Code is widely used because it can be read quickly and in a all round way. However,in terms of the quality of QR Code image, there are always some unavoidable problemscaused by uneven illumination, complicated background, stained bar code and so on,which seriously hampered its further application. In order to solve those problems, thisthesis studied the algorithm of Wavelet Decomposition and Neural Network, andproposed some solutions. First of all, based on the analysis results of the intensityfunction and the reflection function of images, the author proposed to eliminate thenegative impact caused by uneven illumination through wavelet multi scale method todecompose and predict the illumination distribution. And then Otsu’s method is adoptedfor image binarization. In this way decoding is much easier without the poor imagebinarization caused by uneven illumination. Second, the author analyzed the texturalfeatures of QR Code and proposed the information extraction method based on Gaborfiltering and BP neural network. With this method the information stored in the bar codecan be extracted completely and accurately even against complicated background. Theresult shows that the solutions proposed by the author have produced good effect onbinarization and initial image positioning.