Research On Two-dimensional Bar Code Recognition Method

Two-dimensional bar Code has the characteristics of high density and strong reading ability in information storage and transmission.It has been widely used in the fields of item identification,electronic payment,social media,advertising marketing,asset tracking,etc.While being applied in many ways,it also brings problems in recognition technology.A variety of application environments and image acquisition equipment aggravate the complexity of the two-dimensional bar Code recognition process.The traditional two-dimensional bar Code recognition method gradually can't meet the needs of people.In order to improve the robustness and adaptability of the two-dimensional bar Code recognition method in various complex acquisition environments,this paper studies the recognition of the two-dimensional bar Code under the conditions of fuzzy,pollution,damage,occlusion,poor lighting conditions and other harsh conditions.The main work includes the following:(1)When studying the application scenarios and acquisition conditions of 2D bar Code,it is necessary to analyze the complex acquisition environment of 2D bar Code qualitatively and quantitatively,and establish a 2D bar Code image data set as robust as possible under the complex acquisition environment for 2D bar Code training and testing.(2)Referring to and synthesizing a large number of theoretical knowledge analysis,it can be seen that the recognition method of convolutional neural network is usually suitable for training samples;on the contrary,when solving the classification and recognition of small samples and nonlinear problems,support vector machine classifier shows strong robustness.(3)When the bar Code to be tested is identified with different coding databases with a large number of samples,CNN is used.The original image without preprocessing is input to CNN,and the image features are extracted in the convolution,pooling and activation layers,and the extracted feature vectors are mapped to the softmax classifier of the full connection layer to achieve classification.Small convolution kernel is used to complete the convolution process in the experimental simulation process,which avoids the large convolution kernel and causes the difficulty of network separation,resulting in the loss of detailed features.Therefore,CNN can correctly and efficiently realize the identification of the bar Code to be tested and different coding databases.(4)Select a kind of bar Code in different coding database as the sample set,and import the same bar Code to be tested into the sample set,so as to quickly and accurately identify the two-dimensional bar Code which is the same as the template and has complete rules and can be clearly identified.Taking QR Code,the most advantageous and practical QR Code in 2D bar Code database,as an example,a QR Code recognition algorithm based on multi-block local binary patterns and grey wolf algorithm with improved nonlinear convergence factor is proposed to optimize SVM in this paper.The algorithm extracts the MB-LBP features of the high and low frequency components of the block lifting wavelet transform image and fuses them.After the dimension reduction of the fusion features is completed by principal component analysis,the improved GWO optimized SVM classifier is introduced to realize the classification of QR Code data set.The experimental simulation results show that the recognition performance of the improved algorithm is significantly improved.