An image processing apparatus includes a gradient strength calculating unit that calculates a gradient strength of respective pixel values based on an intraluminal image; an extracting unit that extracts a closed region satisfying conditions that the pixel of which gradient strength is a predetermined value or more is not included in the region and a boundary of the region does not bend with a predetermined curvature or higher toward an inner side of the region; and a detecting unit that detects an abnormal portion inside the region. The extracting unit includes a setting unit that sets an initial shape of the region; an energy calculating unit that calculates values of types of energy including at least energies determined respectively by an outer shape of the region and the gradient strength; and an energy-weighted sum calculating unit that calculates a weighted sum of the types of energy.