Detecting with good precision an eye inside corner position and an eye outside corner position as face feature points even when the eye inside corner and/or the eye outside corner portions are obscured by noise. First eyelid profile modeling is performed with a Bezier curve expressed by a fixed control point P3 indicating an eye inside corner first position detected in an image, a fixed control point P4 indicating an eye outside corner first position, a control point P1 corresponding to an upper eyelid position candidate (first parameter), and a control point P2 corresponding to a lower eyelid position candidate (second parameter). Then in a second eyelid profile model with fixed P1 and P2 of the first eyelid profile model having the highest fitting evaluation value λ to the eyelid profile in the image, the values of a control point P3 indicating an eye inside corner position candidate (third parameter) and a control point P4 indicating an eye outside corner candidate (fourth parameter) at a maximum of a fitting evaluation value λ when changing the values of the control point P3 and control point P4 are determined as an eye inside corner second position and an eye outside corner second position, respectively.