In order to solve the difficulty for extraction of roads, buildings and rivers from high-resolution remote sensing images, a new parallel feature detection algorithm based on edge directional information is proposed in this paper. Firstly, parallel edges model is defined as boundaries made up of parallel short lines. Then, the methods of 8-neighborhood boundary-tracking based on collinear restriction in junction points and line detection algorithm in 9-pixels sliding window are proposed to obtain the local directional codes of edge chains. Finally, two efficient criterions are presented to extract the parallel features based on the principal component analysis and the consistency of edge coding. Experimental results show that the proposed algorithm is effective for extracting the nearest parallel lines and curves from high-resolution remote sensing images, and the average accuraty is more than 95%, but the running speed needs further improvement.