Rotationally symmetric vs. oriented frequency channels for the Hotelling observer: a comparison with human observers

This study investigated the inclusion of oriented channels into the Hotelling observer. The performances of the channelized Hotelling observer with rotationally symmetric channels (CHO/sub RS/), the channelized Hotelling observer with rotationally oriented channels (CHO/sub RO/), and human observers were compared for a simple signal detection task. There were four Gaussian signals-circular, elliptical horizontal, elliptical diagonal and elliptical vertical-of equal maximum intensity and signal power. There were four types of lumpy backgrounds-circular, horizontal and vertical-created by filtering a zero-mean, white, Gaussian noise image. For the CHO/sub RO/, the CHO/sub RS/, and the human observers, detectability indices for each of the 16 combinations of signal and background were calculated. The CHO/sub RS/ and CHO/sub RO/ were equivalent for the images where the signal and background had the same orientation; however, the CHO/sub RO/ was more predictive of human performance overall. Neither the CHO/sub RS/ nor CHO/sub RO/ models were predictive of human performance for images of elliptical signals on circularly oriented backgrounds-a result that underscores the need for further refinement of the channel models.