A multi-feed detection apparatus includes a conveyance roller, an ultrasonic sensor for generating an ultrasonic signal, an imaging device to image the medium being conveyed by the conveyance roller and sequentially generate a line image, and a processor to detect a width of the medium in each line image for each line image, detect a vertex count of a medium included in the sequentially generated line image based on the width of the media in the each line image, detect a length of the medium in a direction perpendicular to the medium width, estimate a shape of a medium based on the vertex count and the length of the medium, and detecting media multi-feed by comparing the ultrasonic signal with a predetermined threshold value, and change the predetermined threshold value or stop detection of the medium multi-feed, based on the estimated shape of the medium.