An ideal observer will give equivalent weight to sources of information that are equally reliable. However, when averaging visual information, human observers tend to downweight or discount features that are relatively outlying or deviant ('robust averaging'). Why humans adopt an integration policy that discards important decision information remains unknown. Here, observers were asked to judge the average tilt in a circular array of high-contrast gratings, relative to an orientation boundary defined by a central reference grating. Observers showed robust averaging of orientation, but the extent to which they did so was a positive predictor of their overall performance. Using computational simulations, we show that although robust averaging is suboptimal for a perfect integrator, it paradoxically enhances performance in the presence of "late" noise, i.e. which corrupts decisions during integration. In other words, robust decision strategies increase the brain's resilience to noise arising in neural computations during decision-making.

The discrete wavelet transform (DWT) decomposes an image into bands that vary in spatial frequency and orientation. It is widely used for image compression. Measures of the visibility of DWT quantization errors are required to achieve optimal compression. Uniform quantization of a single band of coefficients results in an artifact that we call DWT uniform quantization noise; it is the sum of a lattice of random amplitude basis functions of the corresponding DWT synthesis filter. We measured visual detection thresholds for samples of DWT uniform quantization noise in Y, Cb, and Cr color channels. The spatial frequency of a wavelet is r 2-lambda, where r is display visual resolution in pixels/degree, and lambda is the wavelet level. Thresholds increase rapidly with wavelet spatial frequency. Thresholds also increase from Y to Cr to Cb, and with orientation from lowpass to horizontal/vertical to diagonal. We construct a mathematical model for DWT noise detection thresholds that is a function of level, orientation, and display visual resolution. This allows calculation of a "perceptually lossless" quantization matrix for which all errors are in theory below the visual threshold. The model may also be used as the basis for adaptive quantization schemes.

The discrete cosine transform (DCT) is a standard method by which images may be compressed. Initially, the image is divided into 8/spl times/8-pixel blocks, each of which is then transformed into its DCT, which is an 8/spl times/8 matrix of coefficients. One image-dependent factor influencing the detectability of DCT basis functions is contrast masking. The authors present data which indicate that sensitivity to a particular coefficient's quantization error is affected by the magnitudes of other coefficients. They propose a revision of Model O to account for between-coefficient contrast masking.< >

The authors have examined the variation in visibility of single DCT basis functions as a function of display visual resolution. They have shown that the existing model (Ahumada and Peterson, 1992; and Peterson et al., 1993) accommodates resolutions of 16, 32, and 64 pixels/degree, provided that one parameter, the peak sensitivity s0, is allowed to vary. Variations in this parameter are to some extent consistent with spatial summation, although sensitivity is lower at the lowest resolution than summation would predict. Practical DCT quantization matrices must take into account both the visibility of single basis functions, and the spatial pooling of artifacts from block to block. Peterson et al. (1993) showed that to a first approximation this pooling is consistent with probability summation. If one considers two images of equivalent size in degrees, but visual resolutions differing by a factor of two, then the sensitivity to individual artifacts would be lower by 4/sup 1/4/ in the higher resolution image due to the smaller block size in degrees, but higher by 4/sup 1/4/ in the same image due to the greater number of blocks. Thus the same matrix should be used with both. The point of the illustration is that the overall gain of the best quantization matrix must take into account both display resolution and image size.< >