Robust Method for E-Maximization and Hierarchical Clustering of Image Classification

We developed a new semi-supervised EM-like algorithm that is given the set of objects present in eachtraining image, but does not know which regions correspond to which objects. We have tested thealgorithm on a dataset of 860 hand-labeled color images using only color and texture features, and theresults show that our EM variant is able to break the symmetry in the initial solution. We compared twodifferent methods of combining different types of abstract regions, one that keeps them independent andone that intersects them. The intersection method had a higher performance as shown by the ROC curvesin our paper. We extended the EM-variant algorithm to model each object as a Gaussian mixture, and theEM-variant extension outperforms the original EM-variant on the image data set having generalizedlabels. Intersecting abstract regions was the winner in our experiments on combining two different typesof abstract regions. However, one issue is the tiny regions generated after intersection. The problem getsmore serious if more types of abstract regions are applied. Another issue is the correctness of doing so. Insome situations, it may be not appropriate to intersect abstract regions. For example, a line structureregion corresponding to a building will be broken into pieces if intersected with a color region. In futureworks, we attack these issues with two phase approach classification problem.