Abstract: This paper presents a multi-scale curvature product corner detection technique in the framework of B-spline curvature scale space. A scale product function is derived from the curvature product of the contour at different scales. Corners are constructed as the local maxima by thresholding the curvature product results across several scales. Through scale product, the localization accuracy and detection performance can be notably improved in terms of CNN criteria. Experiments also demonstrate that proposed method shows robustness to high frequency details and provides promising detection results.