基于RGBD图像色调映射的分割算法研究OA
Study of the Segmentation Algorithm Based on RGBD Image Tone Mapping
针对高背景复杂度、不同光照条件等因素导致的图像分割效果不佳的问题,本文提出了基于RGBD图像色调映射的多特征融合高斯自适应阈值分割算法.首先,利用Modified Mantiuk算法对高动态范围RGBD图像进行色调映射,增强图像的局部区域对比度,并限制对比度增强的程度,以改善图像的背景复杂度指标和光照指标;然后,融合低动态范围图像的8方向L、A、B通道梯度与深度梯度、深度法向量、深度信息等特征,形成多特征融合图;接着,采用基于中心像素点邻域内像素的空间距离接近度与像素融合值相似性机制计算局部阈值,动态判定像素属于前景或背景,进而实现前景分割;最后,通过实验对比验证,与超像素等3种算法相比,该多特征融合高斯自适应阈值分割算法在不同测试集上的准确率达到99.25%,精确率达到98.84%,召回率达到97.46%,F1分数达到98.08%,表现出较强的精度和鲁棒性.
To solve the issue of poor image segmentation results caused by high background complexity and different lighting conditions,the multi-feature fusion Gaussian adaptive threshold segmentation algorithm based on RGBD image tone mapping is proposed.Initially,the improved Mantiuk algorithm is used to perform tone mapping on high dynamic range RGBD images,enhancing the local area contrast of the image and limiting the extent of contrast enhancement,thus improving the image's background complexity and lighting indicators.Subsequently,the eight-directional gradients of the L,A,and B channels from the low dynamic range image are integrated with the depth gradients,depth normal vectors,and depth information features to form a multi-feature fusion map.A local threshold is calculated using a mechanism based on the spatial proximity of pixels within the neighborhood of the central pixel and the similarity of pixel fusion values,to dynamically determine whether they belong to the foreground or background,thereby performing foreground segmentation.Finally,experimental comparisons are conducted.Compared with three algorithms such as superpixel methods,the multi-feature fusion Gaussian adaptive threshold segmentation algorithm achieved an accuracy of 99.25%,a precision of 98.84%,a recall of 97.46%,and an F1 score of 98.08%on different test sets,demonstrating strong precision and robustness.
张兵;詹旭;唐青林
四川轻化工大学自动化与信息工程学院,四川 宜宾 644000四川轻化工大学自动化与信息工程学院,四川 宜宾 644000四川轻化工大学自动化与信息工程学院,四川 宜宾 644000
信息技术与安全科学
RGBD图像色调映射多特征融合前景分割
RGBD imagetone mappingmulti-feature fusionforeground segmentation
《四川轻化工大学学报(自然科学版)》 2026 (1)
35-44,10
四川省科技厅重点研发项目(2022YFS0554)
评论