A Study of Clustering Algorithms and Validity for Lossy Image Set Compression.
Anthony Schmieder, Howard Cheng, Xiaobo Li · 2009
Abstract—A hierarchical lossy image set compression al-gorithm (HMSTa) has recently been proposed for lossy compression of image sets. It was shown that this algorithm performs well when an image set contains well separated clusters of similar images. As a result, if one applies the HMSTa algorithm after a clustering algorithm has been applied, the compression performance depends on the qual-ity of the partition. In this paper, we examine a number of well-known hierarchical clustering methods and cluster validity measures, and their relationships to the compression performance of HMSTa. This relationship can be used as a component in a fully automated image set compression algorithm. We also briefly examine the merit of using different compression schemes depending on the compactness of the cluster in order to reduce computational complexity.