结合改进引导滤波的GrabCut容器前景图像分割

GrabCut Container Foreground Image Segmentation Combined with Improved Guided Filtering

摘要: 针对传统GrabCut算法用户交互后得到的目标容器分割结果存在的边缘凹陷、突刺问题，提出结合引导滤波算法与GrabCut函数的方法改善以上问题。该方法通过用户交互在彩色图像中用矩形框标记出存在容器的目标区域，通过GrabCut算法分割出目标容器，然后结合引导滤波算法，将分割后的结果二值化，作为引导滤波的引导图像掩膜，最后通过引导滤波器结合原图像和引导图像得到目标容器分割结果。实验测量了所提方法与其他对比方法的峰值信噪比(PSNR)、结构相似性(SSIM)及平均运行时间，同其他几种方法相比，实验结果表明，该方法对于GrabCut分割后的边缘凹陷、突刺问题优于对比方法，有明显改善。

Abstract: Aiming at the problem of edge sags and spikes in the segmentation result of the target container obtained after user interaction with the traditional GrabCut algorithm, a method combining the guided filtering algorithm and the GrabCut function was proposed to improve the above problems. This method uses user interaction to mark the target area where the container exists in the color image with a rectangular frame. The target container is segmented by the GrabCut algorithm, and then the guided filtering algorithm is used to binarize the segmented result as a guided image mask for guided filtering Film, and finally the target container segmentation result is obtained by combining the original image and the guided image through a guided filter. The measured peak signal to noise ratio (PSNR), structural similarity (SSIM) and average running time of the proposed method compared with other comparison methods. Compared with several other methods, the experimental results show that this method is effective for GrabCut segmented edges. The problem of sags and spikes is better than the comparative method, and it is significantly improved.