Partial Least Squares Image Clustering
2015 28th SIBGRAPI Conference on Graphics, Patterns and Images
Fig. 1 . A user summary, shown in the first row, represents a possible ground truth. The second row presents a summary obtained with the proposed approach, referred to as PLSIC. Third and fourth rows display summaries obtained by means of our implementation of VSUMM  technique and K-means++, where the number of clusters K is estimated by a shot boundary detection algorithm. The last row is a summary obtained by the original VSUMM method , publicly available. Abstract-Clustering techniques
... have been widely used in areas that handle massive amounts of data, such as statistics, information retrieval, data mining and image analysis. This work presents a novel image clustering method called Partial Least Square Image Clustering (PLSIC), which employs a oneagainst-all Partial Least Squares classifier to find image clusters with low redundancy (each cluster represents different visual concept) and high purity (two visual concepts should not be in the same cluster). The main goal of the proposed approach is to find groups of images in an arbitrary set of unlabeled images to convey well defined visual concepts. As a case study, we evaluate the PLSIC to the video summarization problem by means of experiments with 50 videos from various genres of the Open Video Project, comparing summaries generated by the PLSIC with other video summarization approaches found in the literature. A experimental evaluation demonstrates that the proposed method can produce very satisfactory results.