Region dividing means (3) divides an inputted image by color or edge. Partial feature value providing means (4) extracts visual feature values of the divided region. Area sorting means (5, 6) sorts the feature values of the divided regions and part feature values (32) which are visual features of the parts into which an object is divided in order of area of the region. Part matching means (7) selects the part feature value (32) which the region of the largest area has, selects a visual feature value which the divided region of the largest area has out of the divided regions whose similarities are not calculated, calculates the similarity of the selected visual feature value, and extracts candidate regions. If the visual feature value of the joined region formed by combining the candidate regions is similar to the visual feature value of the whole object stored in storage means (2), the joined region is treated as a region similar to the object.