中文摘要：

为了解决特征学习过程中导致聚类的不均衡性，提出一种基于高斯编码的特征学习算法，使用K—means聚类进行特征训练，在编码过程中考虑了数据分布的影响，同时保留了K-means编码的稀疏性。并且鉴于K-means聚类的不均衡。还提出了一种特征选择的方法用于去噪和降维。改进的模型不仅很大程度上提高了性能而且训练时间和计算代价均小。在人脸数据库AR以及对象分类库Caheeh101上设计了对比实验。实验结果都验证了该算法的有效性和鲁棒性。

英文摘要：

In order to solve the malconformation of clustering in the feature learning, the paper presents a sparse Gaussian coding based feature learning algorithm. It can be trained only through K-means clustering. In the encoding process it takes data＇s distribution into consideration. Given that the K-means clustering often results in unequal clusters, we also propose a feature selection method that can be used for denoising and dimension reduction. This model achieves high accuracy, and saves training time a lot. In this paper,we have designed a contrast experiment on the face database AR and the object database Caltech101. The experimental results show that the algorithm is effective and robust.