Journal of Computer Applications ›› 2020, Vol. 40 ›› Issue (1): 252-257.DOI: 10.11772/j.issn.1001-9081.2019061114

• Virtual reality and multimedia computing • Previous Articles     Next Articles

Multi-exposure image fusion algorithm based on adaptive segmentation

WANG Shupeng, ZHAO Yao   

  1. College of Communication and Information Engineering, Xi'an University of Science and Technology, Xi'an Shaanxi 710054, China
  • Received:2019-06-27 Revised:2019-09-04 Online:2020-01-10 Published:2019-10-08
  • Contact: 王书朋


王书朋, 赵瑶   

  1. 西安科技大学 通信与信息工程学院, 西安 710054
  • 作者简介:王书朋(1975-),男,江苏常熟人,副教授,博士,主要研究方向:图像处理、模式识别、计算机视觉;赵瑶(1995-),女,陕西西安人,硕士研究生,主要研究方向:多曝光图像融合、模式识别。

Abstract: Aiming at the insufficient preservation of color and details existed in traditional multi-exposure image fusion, a novel multi-exposure image fusion algorithm based on adaptive segmentation was proposed. Firstly, the input image was divided into blocks with the same color by super-pixel segmentation. Then structural decomposition was conducted on the image blocks to obtain three individual components. Different fusion rules were designed according to the characteristics of each component, so as to preserve the color and details in original images. Then, the weight map of each component, signal strength component and brightness component were smoothed by guided filtering, effectively overcoming the problem of block effect, retaining the edge information in the source image and reducing the artifacts. Finally, the fusion image was obtained by reconstructing three fused components. The experimental results show that, compared to the traditional fusion algorithms, the proposed algorithm has the average increase of 53.6% in Mutual Information (MI) and 24.0% in Standard Deviation (SD) respectively. The proposed image fusion algorithm can effectively preserve the color and texture details of input images.

Key words: multi-exposure image, image fusion, super-pixel segmentation, fusion rule, guided filtering

摘要: 针对传统多曝光图像融合存在颜色和细节信息保留不完整的问题,提出了一种新的基于自适应分割的多曝光图像融合算法。首先,采用超像素分割将输入图像分割为颜色一致的图像块,再利用结构分解将图像块分解为三个独立分量。根据各分量特点设计不同融合规则,以保留源图像中的颜色和细节信息。然后,采用引导滤波平滑各分量的权重图以及信号强度分量和亮度分量,有效地克服块效应缺陷,保留源图像中的边缘信息,减少伪影。最后,重构融合后的三个分量,得到最终的融合图像。实验结果表明,与传统的融合算法相比,所提算法在互信息(MI)上平均提升了53.6%、标准差(SD)上平均提升了24.0%。该算法能够有效地保留输入图像的颜色和细节纹理信息。

关键词: 多曝光图像, 图像融合, 超像素分割, 融合规则, 引导滤波

CLC Number: