Boosting image classification through semantic attention filtering strategies

In this paper, we present three attention filtering strategies based on their saliency map that improve image classification in three different frameworks: Bag of Visual Words, Spatial Pyramid Matching and Convolutional Neural Networks. These approaches remove significant features at the region level that do not belong to the object of interest. We initially propose AutoBlur, a simple but effective approach to automatically select the blurring factor of the Hou's image signature algorithm used in this paper. Then, based on AutoBlur, we introduce two variants of our approach SARF (Semantic Attention Region Filtering), to semantically remove non-relevant regions through a Mean Shift segmentation. The first is based on the intersection of the Hou's image attention areas with its Mean Shift segmentation, while the second one discards regions based on a key point voting system that uses Euclidean distance.120