|
|
- 2018
基于快速区域建议网络的图像多目标分割算法
|
Abstract:
摘要： 针对传统方法在语义分割中存在大量冗余、结果重叠,造成图像分割算法的结果正确率、鲁棒性较差等问题,提出一种基于快速区域建议网络的图像多目标分割算法。使用选择性搜索(selective search, SS)算法给出初始候选框;采用快速区域建议网络从初始候选框中分类出初始分割框;使用图割算法(GrabCut)从初始分割框中分割出多目标。为了验证本研究算法,采用ImageNet上预训练的VGG16模型,分别使用COCO数据集和CityScapes数据集的训练数据对VGG16模型微调,使用测试数据进行语义分割和多目标图像分割。与YOLO(you only look once,)算法相比,本算法在两个数据集上的平均正确率分别提高了2.16%和1.55%。GrabCut算法在快速区域建议网络的初始分割框上,对多目标的分割更精确,鲁棒性更强。本研究构建的算法通过区域建议网络的得分筛选多目标分割的候选框,保留高得分的候选框来提升图像多目标分割的精度,在多目标的模式识别场合中拥有广泛前景。
Abstract: Aim at the shortage of a large amount of redundancy and overlaps of conventional semantic segmentation algorithm, these shortages caused the image segmentation results getting lower accuracy and robust. A new algorithm of multi-objects image segmentation based on faster region proposal networks was proposed. A selective search algorithm was used to get the initial proposal boxes; a faster region proposal network was used to get initial image segmentation boxes. In order to validate our proposed algorithm, the VGG16 models that pre-trained on ImageNet was used on this problem. By using COCO dataset and Cityscapes dataset, the model was well fine-tuned. The test dataset was used for testing semantic segmentation and image segmentation. Compared with YOLO algorithm, the experimental results showed that our proposed algorithm increased mAP of 2.16% and 1.55%. The initial image segmentation boxes by faster region proposal networks were best fitted by GrabCut, multi-objects segmentation results were more accurate and robust. Our proposed algorithm got higher accuracy by sacrifice little time consumption, which got more application scenes in multi-object patterns recognition
| [1] | 姜枫,顾庆,郝慧珍,等.基于内容的图像分割方法综述[J].软件学报,2017,28(1):160-183. JIANG Feng, GU Qing, HAO Huizhen, et al. An overview of content-based image segmentation methods[J]. Journal of Software, 2017, 28(1):160-183. |
| [2] | 申铉京,张赫,陈海鹏,等.快速递归多阈值分割算法[J].吉林大学学报(工学版),2016,46(2):528-534. SHEN Xuanjing, ZHANG He, CHEN Haipeng, et al. Fast recursive multi-threshold segmentation algorithm[J]. Journal of Jilin University(Engineering and Technology Edition), 2016, 46(2):528-534. |
| [3] | 钱超,张晓林.基于稳定特征点和SLIC超像素分割的快速立体匹配[J].电子设计工程,2016,24(23):146-148,152. QIAN Chao, ZHANG Xiaolin. Fast stereo matching based on stable feature point and SLIC hyperpixel segmentation[J]. Electronic Design Engineering, 2016, 24(23):146-148,152. |
| [4] | REGO Liviane, SUMAILI Jean, MIRANDA Vladimiro, et al. Mean shift densification of scarce data sets in short-term electric power load forecasting for special days[J]. Electrical Engineering, 2017, 99(3):881-898. |
| [5] | LI Z, TANG J. Weakly supervised deep metric learning for community-contributed image retrieval[J]. IEEE Transactions on Multimedia, 2015, 17(11): 1989-1999. |
| [6] | CROMMELINCK S, BENNETT R, GERKE M, et al. SLIC superpixels for object delineation from UAV data[C] //ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Lund, Sweden: Directory of Open Access Journals, 2017(4): 9-16. |
| [7] | LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436-444. |
| [8] | ZOPH Barret,VASUDEVAN Vijay, SHLENS Jonathon, et al. Learning transferable architectures for scalable image recognition[J]. arXiv preprint arXiv:1707.07012(2017). |
| [9] | SCHMIDHUBER J. Deep learning in neural networks: An overview[J]. Neural networks, 2015, 61: 85-117. |
| [10] | 陈海鹏,申铉京,龙建武,等.采用高斯拟合的全局阈值算法阈值优化框架[J].计算机研究与发展, 2016, 53(4): 892-903. CHEN Haipeng, SHEN Xuanjing, LONG Jianwu, et al. The global threshold optimization framework of gaussian fitting is adopted[J]. Journal of Computer Research and Development, 2016, 53(4):892-903. |
| [11] | CORDTS M, OMRAN M, RANMOS S. The cityscapes dataset for semantic urban scene understanding[C] //Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE, 2016: 3213-3223. |
| [12] | NOWOZIN S. Optimal decisions from probabilistic models: the intersection-over-union case[C] //Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Columbus, USA: IEEE, 2014: 548-555. |
| [13] | WANG X, LI H, BICHOT C E, et al. A graph-cut approach to image segmentation using an affinity graph based on l 0-sparse representation of features[C] //2013 20th IEEE International Conference on Image Processing(ICIP). Melbourne, Australia: IEEE, 2013: 4019-4023. |
| [14] | 刘金平,陈青,张进,等.基于集成学习的交互式图像分割[J].电子学报,2016,44(7):1649-1655. LIU Jinping, CHEN Qing, ZHANG Jin, et al. Interactive image segmentation based on integrated learning[J]. Journal of Electronics, 2016, 44(7):1649-1655. |
| [15] | 王春波,董红斌,印桂生,等.基于Hadoop的超像素分割算法[J].计算机应用,2016,36(11):2985-2992. WANG Chunbo, DONG Hongbin, YIN Guisheng, et al. Hyperpixel segmentation algorithm based on Hadoop[J]. Journal of Computer Applications, 2016, 36(11):2985-2992. |
| [16] | YU, SHAODE. Efficient segmentation of a breast in B-mode ultrasound tomography using three-dimensional GrabCut(GC3D)[J]. Sensors,2017,17(8): 1827. |
| [17] | LIN T Y, MAIRE M, BELONGIE S,et al. Microsoft coco: common objects in context[C] //European Conference on Computer Vision. Zurich: Springer International Publishing, 2014: 740-755. |
| [18] | REDMON Joseph, FARHADI Ali. YOLO9000: better, faster, stronger[C] //Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, USA: IEEE, 2017:1384-1392. |
| [19] | REDMON Joseph, DIVVALA Santosh, GIRSHICK Ross, et al. You only look once: unified, real-time object detection[C] //Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE, 2016:3213-3219. |
| [20] | REN Shaoqing, HE Kaiming, ROSS Girshick, et al. Object detection networks on convolutional feature maps[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 39(7): 1476-1481. |
