A medium conveying apparatus includes a conveying roller to convey a rectangular medium, and a processor to generate an input image acquired by imaging the conveyed medium, detect an overlap of the conveyed medium, detect an object area from the input image, determine whether an outer shape of the object area is rectangular, determine whether a multi-feed of the medium has occurred based on whether the overlap of the medium has been detected and whether the outer shape of the object area is rectangular, and execute an abnormality processing when it is determined that the multi-feed of the medium has occurred.