Differences of Face and Object Recognition in Utilizing Early Visual Information

The first cortical stage in both object and face recognition by humans is generally presumed to be a filtering of the image by cells which can be approximated as oriented spatial frequency kernels. Are the outputs of these filters mapped in the same manner to the separate patches of tissue in extrastriate cortex presumed to code faces and objects? Complementary images of objects and faces were produced by dividing the Fourier spectrum of each image into 8 frequency bands and 8 orientation bands. In the inverse Fourier transform, half the 8 × 8 values (analogous to all the red squares of a checkerboard in a row by column representation of frequency and orientation) were contained in one member of a complementary image pair and the remaining combinations of values (e.g., black squares) were contained in the other member. In a naming task original and complementary images produced equivalent priming (equal RTs and error rates) for objects, but name verification for famous faces showed less priming for the complementary image. One possible explanation for these results is that faces are represented as a direct mapping of the outputs of early filter values whereas objects are recognized by means of intermediate primitives (e.g., parts), in which the same primitives can be activated by many patterns of filter activations. Two additional experiments using nonface but highly similar shaped objects (chairs) and unfamiliar faces confirmed the above hypothesis.