Effects of Visual Noise

When an observer is asked to detect a sinewave grating, is squared threshold proportional to the total noise power (including the equivalent noise) passed by a single linear filter ("channel") which depends on the signal, but not on the noise? Both hypotheses will be experimentally tested along several dimensions of spatio-temporal frequency. The hypotheses allow ·~he modulation transfer function (HTF) of the filter to be derived from thresholds measured in bandlimited noise. By bandlimiting the noise in various dimensions the MTF·may be determined over horizontal and vertical spatial frequencies and temporal frequency. In this way, the filter bandwidth (full bandwidth at half amplitude) is found to be about 7.5 c/deg horizo·ntally (for a 5 c/deg horizontal grating), .6 c/deg vertically (for 1, 3, and 9 c/deg horizontal gratings), and 16Hz temporally (for a .25 c/deg grating which was static, or temporally modulated at 4 or 16Hz). CHAPTER 3: NON-WHITE NOISE INTRODUCTION INTRODUCTION Page 88 Chapter 2 found that the squared contrast threshold for many patterns is proportional to the effective spectral density of a white noise background. This chapter will consider how different frequency components of the noise contribute to the threshold elevation of a sinusoidal grating. Chapter 2 was primarily concerned with effects of noise level (of white noise). This chapter examines effect of cut-off frequency and bandwidth (of bandlimited white noise 2 ). The initial hypothesis will be that for any given test pattern there is a linear filter such that the observer's squared threshold is proportional to the total noise power passed by the filter, regardless of the noise spectrum. Later we will acknowledge that the observer may have many such linear filters ~vailable and need not always use the same filter to detect a given pattern. The assumptlons of Theory of Signal Detectability are not satisfied here, so detective efficiency is undefined. However the finding that threshold may be raised by noises whose spectra do not overlap the signal's spectrum is obviously relevant to understanding why the calculation efficiency, in ·white noise, can be so low. Furthermore, it is only by manipulation of noise bandwidth that one can determine at what bandwidth the noise becomes white, i.e. exceeds the bandwidth of the mechanism under study. Several different lines of evidence suggest that human detection of spatial-frequency gratings is mediated by independent mechanisms, "channels", 2 "Bandlimited white noise" is not white noise. White noise has a constant spectral density and frequency components with independent amplitudes over the bandwidth of the mechanism under study. "Bandlimited white noise" is white over a stated band. This band usually will not include the entire bandwidth of the mechanism under study, and thus the noise will not be white. CHAPTER 3: NON-WHITE NOISE INTRODUCTION Page 89 each sensitive over one or two octaves of spatial frequency (on contrast adaptation see Blakemore and Campbell, 1969; on summation see Graham, 1977; on masking by gratings see Legge and Foley, 1980; noise masking will be reviewed below). We will see that effects of visual noise are consistent with this