Context-Aware and Locality-Constrained Coding for Image Categorization

Figure 4

The noise removing functions of 𝑤. We select some categories from Caltech101 and Caltech256 for experiment. Local descriptors are densely extracted with step size of 8 pixels. And the blue circles represent the descriptors preserved under condition 𝑤 > 0.4. As can be seen, most of the background descriptors are removed while the common features of each class are preserved. You may magnify this figure for details.