A base image is created from RGB image signals. A B/G ratio between the B image signal and the G image signal is calculated, and a G/R ratio between the G image signal and the R image signal is calculated. A frequency component-enhanced image in which a frequency component corresponding to an abnormal part different from a normal part is enhanced is generated based on the base image. A composition ratio which indicates the ratio for composing the frequency component-enhanced image with the base image is set based on the B/G ratio and the G/R ratio. The frequency component-enhanced image is composed with the base image at the set composition ratio to generate a structure-enhanced image in which the structure of the abnormal part is enhanced.