Objects with well-defined closed boundary can be discriminated by looking at the norm of gradients. With suitable resizing of their corresponding image windows into a small fixed size(8×8), and further binarized the normed gradients (BING) of images can describe the generic objectness measure. Inspired by the “BING” and considered the character that the artifical targets have many obvious corner points, in this paper we propose to predict candidate windows based on corner points instead of non-maximal suppression the BING used. We can generate a small set of high quality target windows and yield 96.2% object detection rate (DR) like the BING dose but need only half time. This is because of the number of corner point is much less than the number of non-maximal suppression point. Our method generate a small set of high quality target window.