We consider the problem of recovering of continuous multi-dimensional functions from the noisy observations over the regular grid. Our focus is at the adaptive estimation in the case when the function can be well recovered using a linear filter, which can depend on the unknown function itself. In the companion paper "Nonparametric Denoising of Signals with Unknown Local Structure, I: Oracle Inequalities" we have shown in the case when there exists an adapted time-invariant filter, which locally recovers "well" the unknown signal, there is a numerically efficient construction of an adaptive filter which recovers the signals "almost as well". In the current paper we study the application of the proposed estimation techniques in the non-parametric regression setting. Namely, we propose an adaptive estimation procedure for "locally well-filtered" signals (some typical examples being smooth signals, modulated smooth signals and harmonic functions) and show that the rate of recovery of such signals in the $\ell_p$-norm on the grid is essentially the same as that rate for regular signals with nonhomogeneous smoothness.