Currently, we have succeeded in using linear subspace analysis of one feature polar transformation for visual attention detection. But how to relate recursive linear subspace analysis to hierarchical analysis of visual attention. So should we design a new recursive way in linear subspace analysis or should we give a new representation of image data and give the correct meaning of recursive linear analysis?

Key Issue: how to map the dimensionality of subspace to the hierarchical structure of attention (in scale or in feature)?

Initial investigation:
1. Goolgling: linear subspace recursive, linear subspace hierarchical, hierarchical structure attention.