关键词: Hadoop, 并行计算框架（MapReduce）, 图像处理, OpenCV

Abstract: While dealing with huge amount of small image data, Hadoop has the problems of managing the excessive fragmentation of the inputs and saving the rapid growth of small image files. In view of solving these problems, the solution of a new mass small image parallel processing model is proposed and implemented, and is different from the methods such as HIPI and SequenceFile. For Hadoop is suitable for the text-only data processing, the image data is replaced by the text file that stores the image path as input, and the model does not need to design image data types. The functions such as image reading, image processing, image storage are completed in the Map stage of Hadoop. And to simplify the image processing algorithms, the OpenCV functions are combined with the Map function and the corresponding storage method is designed to accommodate the storage of small image files. Experimental results show that, the model has good performance on throughput test and good stability wherever the test data is the small amount of data or large amount of data in Apache Hadoop system.

Key words: Hadoop, MapReduce, image processing, OpenCV