一种新的无监督前景目标检测方法
A New Unsupervised Foreground Object Detection Method
Key words:
foreground object detection, 前景目标检测
unsupervised learning, 无监督学习
feature clustering, 特征聚类
incremental learning, 增量学习
feature extraction 特征提取
Abstract:
Aiming at the low accuracy of object detection methods based on unsupervised object detection (基于无监督特征提取), this paper proposes a foreground object detection method in unlabeled dataset.
针对基于无监督特征提取的目标检测方法效率不高的问题,提出一种在无标记数据集中准确检测前景目标的方法.
The basic idea is that correct(形容词:正确的) feature clustering results can guide(动词,指导) future object feature extraction, while(同时) the accurate foreground object features can improve the accuracy of feature clustering.
其基本出发点是:正确的特征聚类结果可以指导目标特征提取,同时准确提取的目标特征可以提高特征聚类的精度.
The proposed method extracts local features from unlabeled images and then clusters features based on minimum feature distances.
该方法首先对无标记样本图像进行局部特征提取,然后根据最小化特征距离进行无监督特征聚类.
By matching pairwise images in the same cluster,feature weights are computed through feature correspondence.
Finally, the updated feature weights are used to guide feature clustering in the next iteration(迭代).
将同一个聚类内的图像两两匹配,将特征匹配的重现程度作为特征权重,最后,根据更新后的特征权重指导下一次迭代的特征聚类.
We simultaneously group(动词,分组) similar images and detect foreground objects after iterations.
我们同时对相似的图像进行分组,并在迭代后检测前景对象
The experimental results on Caltech256 and Google car side images demonstrate the effectiveness of our method.
实验结果表明,该方法有效地提高Caltech256数据集和Google车辆图像的检测精度
Furthermore, due to the present unsupervised object detection methods lacking of incremental learning ability(增量学习),
we propose an incremental implementation of our method.
此外,针对目前绝大部分(多译)无监督目标检测方法不具备增量学习能力这一缺点(多译),提出了增量学习方法实现,
The experimental results show the incremental learning method can improve the computation speed greatly.
实验结果表明,增量学习方法有效地提高了计算速度.