An approach of extracting reliable feature points alone ridges/troughs is presented. The feature points including high curvature points along ridge/trough,cross points of ridges/troughs and minimum points in image surface. Experimental result shows that the approach is more robust as compare with the traditional approach of extracting feature points along step edges and reduce the possibility of the appearance of outliers. Scanning the image for just one time will get the points along ridges/troughs,but computing the cross points of ridges/troughs and the high curvature points along ridge/trough will cost more time. So this approach is available for applications that require only a small set of highly reliable feature points.