基于改进卷积神经网络的稠密视差图提取方法  

Dense Disparity Map Extraction Method Based on Improved Convolutional Neural Network

作　　者：黄东振 赵沁[1,2] 刘华巍 李宝清[1] 袁晓兵[1] Huang Dongzhen;Zhao Qin;Liu Huawei;Li Baoqing;Yuan Xiaobing(Key Laboratory of Microsystem Technology,Shanghai Institute of Microsystem and Information Technology Chinese AcademyofSciences,Shanghai 201800,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院上海微系统与信息技术研究所微系统技术重点实验室,上海201800 [2]中国科学院大学,北京100049

出　　处：《激光与光电子学进展》2018年第12期201-208,共8页Laser & Optoelectronics Progress

基　　金：微系统技术国防科技重点实验室基金(CXJJ-17S072)

摘　　要：针对现有的卷积神经网络方法所生成的视差图中细节损失严重的问题,提出了在结构上改进的新方法。将原有网络中特征提取部分的4层卷积结构提升到7层,最大化提高了精度;在网络中引入了双金字塔结构,将多尺度降采样信息和特征信息进行了融合,保持了输入图像中的原始细节信息。实验结果表明,改进后网络的错误率从3.029%降到了2.795%,生成的视差图具有更好的连通性。According to the problem of the severe detail loss of the disparity map generated by the current convolutional neural network methods,a structural improvement method is proposed.The 4 layers convolutional structure of the feature extraction part from original network is added to 7 layers to maximize the accuracy.And, the proposed dual pyramid structure is introduced to the network to combine the multi-scale down-sampling information with the feature information,which keeps the details of the original input images.Experimental results show that the error rate of the improved network reduces from 3.029% to 2.795%,and the generated disparity maps have better connectivity.

关 键 词：图像处理 视差获取 深度学习 卷积神经网络 立体匹配 

分 类 号：TN911.73[电子电信—通信与信息系统]