Geometric and arabesque patterns have been used extensively in Persian architecture. Documentation of these patterns is very important for repair and conversation purposes. With recent progresses in automatic object recognition methods, automatic mapping can be used as a fast and low cost method for documentation of cultural heritage monuments. In this paper a novel feature extraction method by means of level sets is used for detecting patterns in Persian architecture. Experimental results suggest that our proposed method is effective and outperforms common edge detection methods, such as Canny.