| 摘要: |
| 针对目前复杂度较大的图像中目标分割速度较慢、显著性边界分割不明确等问题，提出了一种融合改进的FT（Frequency-tuned）显著性检测与Grabcut的图像分割算法。该算法首先通过改进基于频率调谐的FT显著性检测方法得到图像中显著性较高的区域，并利用SLIC（Simple Linear Iterative Clustering）算法对显著图进行预处理得到超像素图，能够有效改善边界的分割效果，然后通过以图论GraphCut算法为基础改进的Grabcut算法建立高斯混合模型。为了提高算法效率，通过聚类以超像素代替原像素，并反复迭代高斯混合模型（Gaussian Mixed Model，GMM）参数，最后利用最大流最小割算法得到最优目标分割结果。实验结果表明所提算法能够更准确更高效率地分割图像中的显著性目标，对高分辨率图像也有很好的适用效果，相比于其他算法在分割精度上提高10%左右，并具有较高的分割效率。 |
| 关键词: 图像分割 显著性检测 超像素 高斯混合模型 |
| DOI： |
|
| 基金项目:河北省自然科学基金项目(E2016202341);河北省高等学校科学技术研究项目(BJ2014013) |
|
| An image segmentation algorithm combining FT saliency map with Grabcut |
| WANG Xiaoyu,YANG Fan,FAN Hairui,WEN Jie,PAN Xuran |
| (School of Electronic and Information Engineering,Hebei University of Technology，Tianjin 300401，China;1.School of Electronic and Information Engineering,Hebei University of Technology，Tianjin 300401，China;2.Tianjin Key Laboratory of Electronic Materials and Devices，Tianjin 300401，China) |
| Abstract: |
| In order to solve the problems of slower target segmentation and ambiguous boundary separation in images with large complexity,an improved image segmentation algorithm based on frequency-tuned(FT) saliency and Grabcut is proposed.The algorithm firstly obtains a region with higher significance in the image by improving the FT saliency detection method based on frequency tuning,then uses the simple linear iterative clustering(SLIC) algorithm to preprocess the significant images to get a superpixel map,which can effectively improve the segmentation effect of the boundary,and then a Gaussian mixture model is built by using the improved Grabcut algorithm based on Graph algorithm.In order to improve the efficiency of the algorithm,the original pixels are replaced by superpixels by clustering,and the Gaussian mixed model(GMM) parameters are iterated repeatedly.Finally,the optimal target segmentation result is obtained by using the maximum flow minimum cut algorithm.The experimental results show that the algorithm can more accurately and efficiently segment significant targets in images and has a good effect on high resolution images.The proposed segmentation method achieves better segmentation efficiency and its segmentation accuracy has increased by about 10% on average in comparison with that of other existing algorithms. |
| Key words: image segmentation significant detection superpixel Gaussian mixed model(GMM) |
