Attention and biased competition in multi-voxel object representations