OTLines: A novel line-detection algorithm without the interference of smooth curves

We propose a novel line-detection algorithm that avoids interpreting a smooth curve in an image as many straight segments. The algorithm starts from a contiguous (connected) set of edge chains extracted by the Edge Drawing Parameter Free (EDPF) algorithm such that a problem of line detection in the image space is transformed to a problem of line segments and circular arcs identified for each edge chain. The core of our approach is the proposed orientation transformation (OT) algorithm based on principal component analysis; hence, our algorithm is named OTLines. With the OT algorithm, each edge chain is not only decomposed into a series of straight line segments and smooth curve candidates at the estimated corners but is also mapped to a series of oblique lines and horizontal lines in a new space called the orientation space. Based on the results of the transformation, line segments and circular arcs can be detected simultaneously in both the orientation and image spaces using principal component analysis. The performance of our proposed algorithm is validated using a set of typical images. We also compare our algorithm with existing algorithms. Our novel algorithm is very effective at segmenting the lines in synthetic and natural images. No parameters in the proposed algorithm must be tuned during its application.