Training-Free Attentive-Patch Selection for Visual Place Recognition