A High-performance Computing Method for Photographic Mosaics upon the Hadoop Framework

Abstract

Since digital images are non-structure data, in order to apply some image processing techniques on digital images, we need high computing power. Nowadays, digital images have become an issue of Big Data, so we decide to implement an adaptive K-Medoids based on a popular Big Data analysis tool, Hadoop, in our research. Hadoop can provide high computing power to do processes of high computational complexity required by the algorithm of mosaic images, especially, in a Big Data environment where there are tons of images to be dealt. Our research focuses on three main goals. First, we use an unsupervised clustering method, K-Medoids, to cluster the image dataset and build a codebook, and then we can use the codebook to generate the mosaic image to reduce the processing time. Second, we use two feature selection metrics to develop an adaptive K-Medoids method, called feature-based K-Medoids (FKM), which can cluster the image dataset faster by the feature selection mechanism. Third, our method surely reduces the processing time of mosaic images by the codebook. Though the image quality by our method is slightly lower compared with Szul et al.’s method, our method retains an acceptable image quality.