基于非归一化直方图的GrabCut图像分割算法改进

针对GrabCut算法在图像分割中存在迭代求解耗时长、分割结果欠分割的问题，提出了一种基于非归一化直方图改进的GrabCut算法。在保留GrabCut第一次分割结果的基础上，通过非归一化直方图计算像素点属于前景或背景的方法来代替高斯混合模型迭代学习的过程；在构图过程中引入一类新的节点Bin进行构图以提高分割精度。选取MSRA1000数据集中部分图片进行实验验证，结果表明该算法在分割效果和效率上都有明显的提升，在进行背景复杂图像的分割时改进算法优势更加明显。

Improvement of GrabCut image segmentation algorithm based on non-normalized histogram

Aiming at the time efficiency issue and segmentation effect problem of GrabCut image segmentation algorithm, this paper proposed an improved GrabCut algorithm which based on non-normalized histogram. The results of the first segmentation of Grabcut were preserved, and in subsequent iterations learning process of Gaussian mixture model was replaced by the non-normalized histogram calculation of pixels belonging to foreground or background; A new type of node Bin was introduced in the composition step to compose the image to improve the segmentation precision. Experiment results on MSRA1000 data set indicate that the algorithm has a significant improvement in both segmentation effect and efficiency. The advantage of the improved algorithm is more obvious when the complex background image is being segmented.