文献类型：会议论文

英文题名：Exposure fusion based on sparse coding in pyramid transform domain

作者：Wang, Jinhua[1]; Xu, Guangmei[1]; Lou, Haitao[1]

第一作者：王金华

机构：[1] Institute of Information Technology, Beijing Union University, Beijing, 100101, China

第一机构：北京联合大学智慧城市学院

会议论文集：ICIMCS 2015 - Proceedings of the 7th International Conference on Internet Multimedia Computing and Service

会议日期：August 19, 2015 - August 21, 2015

会议地点：Zhangjiajie, Hunan, China

语种：英文

外文关键词：Computation theory - Image coding - Laplace transforms - Textures

摘要：Sparse representation theory explores the sparseness of natural signals, and can be used to fuse multi-exposure images. To reduce the presence of artifacts, a "sliding window" technique with a small step size is usually adopted, enabling high-quality images to be obtained. However, this approach is time-consuming, especially when the source image is large (i.e., 1280×1024 pixels). The method proposed in this paper reduces the computation time of the image fusion process. We know that the low-frequency component gives a coarse representation of the original image, and inherits certain properties such as the mean intensity and texture information. First, we obtain the low-frequency subimage using Laplacian pyramid decomposition of the source images. These low-frequency components are used to represent the original image, and are fused using the sparse coding fusion framework. Second, we apply saliency matching as a fusion rule for the high-frequency subimage obtained by Laplacian pyramid decomposition. Experiments show that the proposed method produces comparative or even better results than other typical exposure fusion methods, and greatly reduces the computation time. ? 2015 ACM.