A new algorithm to feature detection

This paper describes a new feature detection algorithm, which bases on color and curvature. Since color provides valuable information in object recognition and description, a color invariant space is adopted instead of the gray space. With this method, the distinctive features extracted from images are invariant to image scale and rotation. To reduce the computational complexity and improve the quality of detection, the cascade filtering approach is employed. Details of the new feature descriptor are described, along with test results.