The role of gist in scene recognition

Studies of change blindness suggest that we bring only a few attended features of a scene, plus a gist, from one visual fixation to the next. We examine the role of gist by substituting an original image with a second image in which a substitution of one object changes the gist, compared with a third image in which a substitution of that object does not change the gist. Small perceptual changes that affect gist were more rapidly detected than perceptual changes that do not affect gist. When the images were scrambled to remove meaning, this difference disappeared for seven of the nine sets, indicating that gist and not image features dominated the result. In a final experiment a natural image was masked with an 8x8 checker pattern, and progressively substituted by squares of a new natural image of the same gist. Spatial jitter prevented fixation on the same square for the sequence of 12 changes. Observers detected a change in an average of 2.1 out of 7 sequences, indicating strong change blindness for images of the same gist but completely different local features. We conclude that gist is automatically encoded, separately from specific features.