1. What are the contributions in "Mining mid-level features for image classification" ?
In this paper the authors propose a new and effective scheme for extracting mid-level features for image classification, based on relevant pattern mining.. The authors show that these bag-of-FLHs are more discriminative than traditional bag-of-words and yield state-of-the-art results on various image classification benchmarks, including Pascal VOC.
read more
2. What are the future works mentioned in the paper "Mining mid-level features for image classification" ?
As future work the authors propose to investigate how to push the relevant and non redundant constraints directly into the local histogram mining process to make it particularly efficient.. The authors also plan to further investigate pattern mining methods in image classification, image/video retrieval and multiple query image retrieval settings.. Furthermore, the authors would like to further explore unsupervised relevant pattern mining techniques and how to extend FLH using Gaussian mixture-based visual word representations and probabilistic mining.. Besides, as itemset mining is performed on the local histograms, the spatial information that can be captured in the patterns may be limited in some applications.
read more
3. What is the effect of the binary histogram transformation on the FIM method?
Since the binary histogram transformation used in the FIM method loses some valuable information, this method does not benefit that much from the relevant pattern selection step nor from the reduction of the dictionary size, SIFT dimension and the increase of the spatial neighborhood.
read more
4. What is the effect of smaller dictionaries on FLH?
The use of smaller dictionaries and less discriminative SIFT features allows the FLH -based method to exploit larger neighborhoods.
read more
