A Robust and Efficient Feature Extract Transform

The efficient feature detection is a crucial step for various tasks in computer vision. However, the corner location is not very accurate and some edge pixels have high responses for the kind of corner detector based on the accelerated segment test. Furthermore, this kind of method is sensitive to the change of illumination. This paper presents a robust and efficient feature extraction transform (REFET). The method includes three steps: firstly, the input image is enhanced based on the color constancy of the human vision system. Then, acoarse edge pixel detection method is used to eliminate edge points in the image. Finally, an adaptive mask is proposed to extract corners accurately and efficiently. The experiments demonstrate that the proposed method improves the corner localization accuracy and the robustness for illumination, and the edge points are suppressed in the results.