A new nonlinear intensity interpolation algorithm is presented to realize sub-pixel edge detection. The interpolation algorithm based on the Canny criteria makes full use of grads information attained by Canny edge detection to perform special interpolation in the grads direction. When the resolution is enhanced, the interpolated image by the new interpolation scheme can efficiently preserve high frequency component in the original image. The edge detection of interpolated image permits high precision localization. The new interpolation algorithm is more effective in reserving the grads information of the step edge of the initial image than the usual linear interpolations. It requires simpler computation than the present non-linear interpolations.