摘要： 为提高目标识别的精确度和速度,提出一种模板选取算法。利用训练样本图像的标记信息,使属于不同类的局部神经反应尽量分开,属于同一类的局部神经反应尽量靠近,以减少模板集合中的冗余,并得到数目较少且具有较强判别能力的模板。实验结果表明,与随机选取算法相比,该模板选取算法可以在保证精确度的前提下明显提高目标识别的效率。

关键词: 目标识别, 特征提取, 局部编码, 描述子, k-means聚类算法, 模板选取

Abstract: In order to improve the accuracy and speed of object recognition,a template selection algorithm is presented.By using the labeling information of training sample images,this algorithm makes the different kinds of local neural response separate and the same ones come close.It not only reduces the redundancy of the template sets,but also obtains a small number of templates with strong discrimination ability.Experimental results show that,compared with the random selection algorithm,the proposed algorithm can significantly improve the efficiency of object recognition under the premise of ensuring accuracy.

Key words: object recognition, feature extraction, local coding, descriptor, k-means clustering algorithm, template selection

中图分类号: 

• TP301.6