|
| 基于自适应增强与混合特征的低照度图像拼接方法 |
| Low-Brightness Image Splicing Algorithm Based on Improved SIFT Hybrid Features |
| 投稿时间：2026-06-16 修订日期：2026-07-14 |
| DOI： |
| 中文关键词: 低照度图像 图像增强 多尺度Retinex 特征匹配 图像拼接 |
| 英文关键词: Low-light images Image enhancement Multi-scale Retinex Feature matching Image stitching |
| 基金项目:贵州省科技计划项目(黔科合人才KJZ[2025]084)； |
|
| 摘要点击次数: 128 |
| 全文下载次数: 0 |
| 中文摘要: |
| 低照度图像拼接在夜间监控、无人系统视觉感知及弱光环境三维重建等场景中具有重要应用价值。然而受光照条件限制,低照度图像普遍存在亮度不足、对比度偏低及特征点稀疏等问题,导致传统拼接方法在特征提取与匹配过程中稳定性较差,易出现匹配失败与拼接失真。针对上述问题,提出一种基于自适应增强与混合特征的低照度图像拼接方法。该方法通过融合特征点密度与局部信息熵构建自适应增强权重,并引入梯度一致性约束,实现图像亮度恢复与结构信息协同优化；在此基础上,设计混合特征表达机制,提高低照度条件下特征描述的稳定性与匹配可靠性。实验结果表明,该方法在增强效果与拼接质量方面均优于传统方法,其中信息熵显著提升,匹配内点率达到96.7%,拼接结果PSNR与SSIM分别达到13.58和0.862。与传统SIFT算法相比,所提方法在鲁棒性与拼接性能方面具有明显优势。 |
| 英文摘要: |
| Low-light image stitching is of considerable significance in applications such as nighttime surveillance, autonomous visual perception, and three-dimensional reconstruction under insufficient illumination. However, due to the inherent limitations of low-light imaging conditions, acquired images generally exhibit low brightness, reduced contrast, and sparse feature distributions. These characteristics degrade the stability of feature extraction and matching in conventional stitching methods, frequently resulting in mismatches and geometric distortions.To address these challenges, this paper proposes a low-light image stitching method based on adaptive enhancement and hybrid feature representation. Specifically, adaptive enhancement weights are constructed by integrating feature point density with local information entropy, enabling differentiated enhancement across image regions. Furthermore, a gradient consistency constraint is incorporated to preserve structural information during the illumination enhancement process, thereby achieving a balance between brightness restoration and detail retention. On this basis, a hybrid feature representation mechanism is developed to improve the robustness of feature descriptors and the reliability of feature matching under low-light conditions.Experimental results demonstrate that the proposed method achieves superior performance in both enhancement quality and stitching accuracy. The information entropy is significantly increased, the inlier matching rate reaches 96.7%, and the PSNR and SSIM of the stitched images attain 13.58 and 0.862, respectively. Compared with the conventional SIFT-based method, the proposed approach exhibits clear advantages in terms of robustness and overall stitching performance. |
|
View Fulltext
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
