A method including normalizing an original image into an intermediate image, the intermediate image including multiple local blocks; calculating image feature data of the local blocks; calculating weight distribution data corresponding to the local blocks in the intermediate image according to the image feature data, the weight distribution data representing a degree of possibility that the local blocks include part or all of an object; and determining a location area of the object in the original image based on the weight distribution data obtained by calculation. By using the technical solutions in this present disclosure, an object in an image is localized rapidly and efficiently, and a subject area is determined, thereby saving a large amount of work for manually labeling images.