In this paper, we propose and evaluate a method for automatic frame extraction from a collection of historical photographs. These frames are very noisy and were demonstrated to significantly affect the results of content-based image indexing and retrieval in the photograph images. The method is based on parallelogram detection that uses a Hough transform variation called Tiled Hough Transform in which the image is split into tiles to reduce computational complexity. This detector is then extended to combine detected parallelograms into a resulting frame. Two key contributions of this work are: (1) a new effective technique to solve the photographs frame problem, and (2) the use of a set of statistical and experimental design techniques either to fine-tune the proposed method and to demonstrate its effectiveness