CAJ | 학술논문

复杂背景下的二维条码区域提取一直是Data Matrix条码解码过程中的难题之一.通过对图像进行形态学分析从而确定条码的可能区域的方法因计算简单而被广泛应用.但存在着形态学结构体难以选择和虚警率比较高的缺点.为克服这些缺点,提出了基于Gabor滤波和BP神经网络的Data Matrix条码区域提取方法(GF-BPNN).用不同尺度不同方向的Gabor滤波器对图像进行滤波提取其纹理特征,再进行特征变换,使所得特征具有尺度和旋转不变性;然后利用BP神经网络按照前述特征对像素进行分类,再经过形态学后处理提取条码区域.实验结果表明,与进行形态学分析的方法相比,GF-BPNN具有较高的准确率和鲁棒性.
복잡배경하적이유조마구역제취일직시Data Matrix조마해마과정중적난제지일.통과대도상진행형태학분석종이학정조마적가능구역적방법인계산간단이피엄범응용.단존재착형태학결구체난이선택화허경솔비교고적결점.위극복저사결점,제출료기우Gabor려파화BP신경망락적Data Matrix조마구역제취방법(GF-BPNN).용불동척도불동방향적Gabor려파기대도상진행려파제취기문리특정,재진행특정변환,사소득특정구유척도화선전불변성;연후이용BP신경망락안조전술특정대상소진행분류,재경과형태학후처리제취조마구역.실험결과표명,여진행형태학분석적방법상비,GF-BPNN구유교고적준학솔화로봉성.
Extracting 2D barcode region in complex backgrounds has always been a key problem in the procedure of Data Matrix barcode decoding.Thanks to its low computation complexity,morphological analysis was widely applied in extracting potential regions of Data Matrix barcodes,with inherent defects of high false accept rate and difficulty in choosing appropriate structuring element though.In order to overcome these drawbacks,the method of Data Matrix barcode region extraction,based on Gabor filtering and BP neural network(GF-BPNN),was proposed.GF-BPNN firstly filtered the image by a set of Gabor filters with different scales and orientations to extract its texture features,and then transformed these features to make them scale and rotate invariant.After that,GF-BPNN employed the BP neural network to classify pixels according to the features aforementioned. Finally, Data Matrix barcode regions were extracted by morphological post-processing.Experiment results revealed that GF-BPNN Was more accurate and robust than morphological analysis.