Plant leaf images segmentation via graph diffusion process

We present the segmentation results obtained from our graph-based diffusion process using random walk with restart on a mono-layered graph using the public Pl@ntleaves (H. Go¨eau, P. Bonnet, A. Joly, N. Boujemaa, D. Barth´el´emy, J.-F. Molino, P. Birnbaum, E. Mouysset, M. Picard, The clef 2011 plant images classification task, Vol. 1177, 2011.) dataset. The dataset comprises 233 high-resolution leaf images captured in their natural environment. The images include various artefacts that pose challenges to the segmentation task, such as shadows, varying illumination, and the presence of overlapping leaves. Our algorithm emphasizes the leaf parts by diffusing intensity scores from foreground templates towards image boundaries. The resulting saliency maps are further refined through a fusion process with saliency maps generated by random forests. The refined saliency maps are then thresholded to extract the leaves from their backgrounds. Ground truth images are available to visually evaluate the effectiveness of our algorithm's performance. Folders description: * JPEGimages: Leaf color images. * masks: The ground truth binary masks that accurately delineat the leaf regions. * foreground_template: contains the bounding boxes that localize the leaves in blue and the foreground templates in red drawn on dataset images. * DF_sal: contains the saliency maps derived from the diffusion process within the graph. * RF_sal: contains the saliency maps generated by random forests. * final_sal: contains the final saliency maps obtained after the fusion process. * PRE_segmentation: contains the segmentation results obtained after thresholding the final saliency map and before refinement. * final_segmentation: contains the final segmentation results obtained after refinement. *SLG_Segmentation results: a compressed folder containing the above folders

本研究展示了基于图的扩散过程（Graph-based Diffusion Process）所得的分割结果，该过程在单层图（Mono-layered Graph）上采用重启随机游走（Random Walk with Restart）方法，所用公开数据集为Pl@ntleaves（H. Goeau, P. Bonnet, A. Joly, N. Boujemaa, D. Barthélemy, J.-F. Molino, P. Birnbaum, E. Mouysset, M. Picard, The clef 2011 plant images classification task, Vol. 1177, 2011.）。
本数据集包含233张在自然环境中采集的高分辨率叶片图像。这些图像存在多种对分割任务构成挑战的图像伪影，例如阴影、光照变化以及叶片重叠现象。我们提出的算法通过从前景模板向图像边界扩散强度分值，以突出叶片区域。所得到的显著性图（Saliency Map）还将通过与随机森林生成的显著性图进行融合处理，得到进一步优化。随后对优化后的显著性图进行阈值处理，以将叶片从背景中分割出来。本次研究提供了真值图像（Ground Truth），用于直观评估所提算法的性能表现。
文件夹说明：
* JPEGimages：叶片彩色图像
* masks：用于精准勾勒叶片区域的真值二值掩码
* foreground_template：包含绘制在数据集图像上的蓝色叶片定位边界框与红色前景模板
* DF_sal：包含基于图内扩散过程得到的显著性图
* RF_sal：包含由随机森林生成的显著性图
* final_sal：包含经融合处理后得到的最终显著性图
* PRE_segmentation：包含对最终显著性图进行阈值处理后、优化前的分割结果
* final_segmentation：包含经优化后得到的最终分割结果
* SLG_Segmentation results：包含上述所有文件夹的压缩文件夹