Image Cataloging and Partitioning Using Hierarchical Conditional Random Field Model
The partitioning (Segmenting) and cataloging (labeling) images is a fundamental problem in Computer Vision. The hierarchical Conditional Random Field model deal with the problem of labeling images by object. When labeling a new image, select the cluster and use the associated CRF model to label this image. Given a test image, one first use the CRF model to obtain initial labels then find the cluster of the image. Finally, re-label the image by the CRF model associated with this cluster. To effectively compare and extract similar images, introduce a new image descriptor, the label-based descriptor which summarizes the semantic information of a labeled image.