基于最优分割的矩匹配图像非均匀性校正方法研究
DOI:
CSTR:
作者:
作者单位:

1.重庆光电技术研究所;2.西安电子科技大学集成电路学部;3.电子科技大学,集成电路科学与工程学院,重庆光电技术研究所;4.集成电路与微系统全国重点实验室;5.集成电路科学与工程学院,电子科技大学;6.集成电路学部集成电路学部,西安电子科技大学

作者简介:

通讯作者:

中图分类号:

基金项目:

国家自然科学基金青年科学基金项目(No.12305311); 重庆市自科基金面上项目(CSTB2024NSCQ-MSX1189)


Research on Image Non-Uniformity Correction Method Based on Optimal Segmentation Moment Matching
Author:
Affiliation:

1.Faculty of Integrated Circuits, Xidian University;2.National Key Laboratory of Integrated Circuits and Microsystems;3.School of Integrated Circuit Science and Engineering, University of Electronic Science and Technology of China;4.Chongqing Optoelectronics Research Institute;5.Faculty of Integrated Circuits, Xidian University, Xian

Fund Project:

The National Natural Science Foundation of China (General Program, Key Program, Major Research Plan)

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对因图像传感器像元固有缺陷导致的条纹噪声严重影响成像质量,以及传统定标法需反复标定、全局矩匹配易过度平滑丢失细节等问题,同时为了在有限的硬件资源平台上实现图像的实时处理,基于最优分割方案,本文设计了一种基于最优分割的矩匹配非均匀性校正方法。首先对输入图像进行列特征预处理,再通过Fisher算法对图像进行保序分块,随后对得到的初始列块进行相似块合并优化,采用查并集算法将差值小于设定阈值的列块合并,最后对每个列块执行矩匹配校正,得到输出图像。实验结果表明,本算法将平场图像的PRNU由4.86%降至0.80%、DSNU由1.73%降至0.77%;对叠加列条纹的标准细节图像,PSNR由22.48 dB提升至33.21 dB,较两点法(31.55 dB)和全局矩匹配(26.17 dB)分别提高1.66 dB和7.04 dB;并给出了分割段数k与两个阈值的实验论证与敏感性分析结果,在真实传感器采集图像上完成了验证。所提方法可有效降低光响应非均匀性,且对原始图像的细节有良好的保真特性,算法易于移植和硬件实现,满足小面积低功耗的要求。

    Abstract:

    To address severe stripe noise from inherent pixel defects in image sensors and the limitations of conventional correction approaches including frequent recalibration and detail loss induced by over-smoothing in global moment-matching, this paper presents a non-uniformity correction method based on optimal-segmentation moment matching for real-time processing under limited hardware resources. Initially,the input image is preprocessed via column feature analysis, followed by order-preserving blocking via the Fisher algorithm. Similar column blocks with differences below the threshold are merged using the union-find algorithm, and moment-matching correction is executed on each block to generate the corrected output. Experimental results show that the proposed method reduces the PRNU of the flat-field image from 4.86% to 0.80% and the DSNU from 1.73% to 0.77%; for a standard detail image with column stripes, the PSNR is improved from 22.48 dB to 33.21 dB, which is 1.66 dB and 7.04 dB higher than the two-point method (31.55 dB) and global moment matching (26.17 dB), respectively. The selection of the segment number k and the two thresholds is further justified by the experimental verification and sensitivity analysis, and the method is verified on images captured by a real image sensor. It effectively suppresses photo-response non-uniformity while preserving image details, and features favorable portability and hardware-friendliness for small-area, low-power scenarios.

    参考文献
    相似文献
    引证文献
引用本文
分享
相关视频

文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2026-09-04
  • 最后修改日期:2026-09-14
  • 录用日期:2026-09-14
  • 在线发布日期:
  • 出版日期:
文章二维码