并行计算在词袋模型算法中的应用

传统词袋模型已广泛地应用于图像处理领域,并取得较好效果.但在传统词袋模型中,仅考虑了串行计算,使得整个算法流程耗时较长.考虑现有的多核CPU资源,结合共享存储并行编程(OpenMP)并行框架,对词袋模型进行并行优化,并对其性能进行讨论.主要考虑对特征提取、特征聚类和图像直方图生成三个部分进行并行优化.通过对Caltech 100数据库进行实验,结果表明,该方法可以取得接近于CPU核数的加速比,因此减少了词袋模型的构造和图像直方图生成时间,相对于传统词袋方法提高了算法的效率.

The application of parallel computing in the bag of words model algorithm

Traditional bag of words (BoW) model has been widely used in image classification,and achieves better results.But in the traditional bag of words model only the serial computation is considered,making the whole process takes a long time.In this paper the BoW model is optimized with parallel method combining with OpenMP parallel framework under multi-core CPU resources.Parallel optimization of three parts,including feature extraction,feature clustering and image histogram generation,is performed.The proposed method is executed on the Caltech 100 database.The experimental results show that the proposed method can achieve speedup ratio which is close to the number of CPU cores.Therefore,compared with the traditional BoW model,the time for vocabulary construction and image histogram generation has been reduced,which improves the algorithm efficiency.