On what basis do you claim that sparse non-orthogonal bases are more likely than Fourier bases? Do they better predict the illusory grouping data, or is this just a personal prejudice of yours?