I haven't tested this, but the package https://github.com/subokita/Robust-Text-Detection apparently implements methods described in the paper "Robust Text Detection in Natural Images with Edge-Enhanced Maximally Stable Extremal Regions." by Chen, Huizhong, et al. (http://web.stanford.edu/~hchen2/papers/ICIP2011_RobustTextDe...), which sounds like it might be the sort of thing you want.

In particular, the figure on the first page of the paper shows detection (not OCR) of a textual logo in a photo.