An image recognition apparatus includes an objective image data acquisition unit to acquire objective image data containing a face image portion, a feature area detection unit to detect an eye area corresponding to left and right eyes contained in the objective image data, an inclination estimation unit to estimate inclination of the face image portion in a depth direction thereof based on the relative position of the detected eye area, an area-to-be-extracted determination unit to determine a shape of an area to be extracted in response to the inclination estimated by the inclination estimation unit, and an area-to-be-extracted extraction unit to extract the area to be extracted of the shape determined from the objective image data. Preferably, the area to be extracted is subjected to predetermined processing.