A region extraction apparatus includes an image acquisition device that acquires an image, a temporary initial region specifying device that specifies a plurality of temporary initial regions in the acquired image, a separation calculation device that calculates a separation that is an indicator indicating how much pixel values of all pixels of an inside of a specified temporary initial region and pixel values of all pixels of an outside of the specified temporary region are different from each other, on each temporary initial region, an initial region specifying device that specifies a temporary initial region having a highest calculated separation as an initial region, and a region extraction device that performs a region extraction on a basis of the specified initial region.