The improved (2D)2PCA algorithm and its parallel implementation based on image block

Both the one-dimensional method based on vector and the two-dimensional method based on matrix in image feature extraction are only suitable for the processing of small scale images. In this paper, we improved the (2D)2PCA algorithm based on image block. In addition to the conventional correlation method, we proposed three methods: image enhancement processing method, image block method and the method of creating projection matrix based on local mean and standard deviation. We also presented the executive process of the (2D)2PCA algorithm on the MapReduce programming model and compared its performance in terms of the stand-alone machine and cluster environment. The experiment results showed that the (2D)2PCA algorithm based on image block has higher execution efficiency, better speedup, better scalability and is suitable for massive image data processing.