电光与控制, 2017, 24 (7): 28, 网络出版: 2017-09-21
基于自适应弹性网络稀疏编码的目标识别
Object Recognition Based on Adaptive Elastic Net Sparse Coding
目标识别 尺度空间 自适应弹性网络 稀疏编码 object recognition AGAST AGAST scale space adaptive elastic net sparse coding
摘要
传统稀疏编码模型特征选择能力较弱, 稀疏系数向量中负系数的存在导致维数偏高、信息冗余, 不利于目标识别。针对这个问题, 提出了一种基于自适应弹性网络的稀疏编码模型。该模型首先利用融合尺度空间的AGAST检测子提取特征点, 经过FREAK算子描述, 采用能够自适应选择强相关性特征的自适应弹性网络回归模型求解稀疏系数向量, 最后通过分类器实现对目标的分类识别。实验结果表明,特征检测算法对于图像中尺度、视角、光照和旋转等变换具有更强的鲁棒性, 在自适应弹性网络的约束下, 模型具有较好的识别性能。
Abstract
The traditional sparse coding model has poor feature selection performance,and the negative coefficient in the sparse-coefficient vectors may cause high dimensions and information redundancy,which is harmful for the object recognition.To solve the problem,a sparse coding model based on the adaptive elastic net is proposed.The model firstly extracts the feature points by AGAST (Adaptive and Generic Corner Detection Based on the Accelerated Segment Test) detector in scale space,and describes them by FREAK algorithm.Then the sparse-coefficient vectors are calculated out by applying an adaptive elastic net regression model that can select strong correlation features.Finally,target recognition is realized by classifier.The result shows that feature detection algorithm is more robust to the changes of scale,viewpoint,brightness,and rotation,and the recognition performance of the model has great improvement under the restriction of adaptive elastic net model.
杜玉龙, 李建增, 张岩, 范聪. 基于自适应弹性网络稀疏编码的目标识别[J]. 电光与控制, 2017, 24(7): 28. DU Yu-long, LI Jian-zeng, ZHANG Yan, FAN Cong. Object Recognition Based on Adaptive Elastic Net Sparse Coding[J]. Electronics Optics & Control, 2017, 24(7): 28.