Selective internal operations in the recognition of locally and globally point-inverted patterns

Performance in discriminating rotated 'same' patterns from 'different' patterns may decrease with rotation angle up to about 90 degrees and then increase with angle up to 180 degrees. This anomalously improved performance under 180 degrees pattern rotation or point-inversion can be explained by assuming that patterns are internally represented in terms of local features and their spatial-order relations ('left of', 'above', etc.), and that, in pattern comparison, an efficient internal sense-reversal operation occurs (transforming 'left of' to 'right of', etc.). Previous experiments suggested that local features and spatial relations could not be efficiently separated in some pattern-comparison tasks. This hypothesis was tested by measuring 'same-different' discrimination performance under four transformation: point-inversion 1 of the whole pattern, point-inversion 1F of local features alone, point-inversion 1P of local-feature positions alone, and identity transformation Id. The results suggested that internal sense-reversal operations could be applied selectively and efficiently, provided that local features were well separated. Under this condition performances for 1F and 1 were about the same whereas performance for 1P was significantly worse, the latter performance resulting possibly from an attempt to apply internal global and local sense-reversal operations serially.