基于稀疏表示全局字典学习的图像分类方法

针对传统的稀疏表示字典学习图像分类方法在大规模分布式环境下效率低下的问题,设计一种基于稀疏表示全局字典的图像学习方法。将传统的字典学习步骤分布到并行节点上,使用凸优化方法在节点上学习局部字典并实时更新全局字典,从而提高字典学习效率和大规模数据的分类效率。最后在MapReduce平台上进行并行化实验,结果显示该方法在不影响分类精度的情况下对大规模分布式数据的分类有明显的加速,可以更高效地运用于各种大规模图像分类任务中。

Image classification based on global dictionary learning method with sparse representation

To address the problem of low efficiency for traditional massive image classification, a sparse representation based global dictionary learning method was designed. The traditional dictionary learning steps were distributed to parallel nodes, local dictionaries were first learnt in local nodes and then a global dictionary was updated in real time by those local dictionaries and variables through using convex optimization method, thereby enhancing the efficiency of dictionary learning and classification of massive data. Experiments on the MapReduce platform show that the new algorithm has better performance than classical image classification methods without affecting the classification accuracy, and the new algorithm can be widely used in massive and distributed image classification tasks.