Perceptual Classification Images from Vernier Acuity Masked by Noise

Letting external noise rather than internal noise limit discrimination performance allows information to be extracted about the observer's stimulus classification rule. A perceptual classification image is the correlation over trials between the noise amplitude at a spatial location and the observer's responses. If, for example, the observer followed the rule of the ideal observer, the response correlation image would be an estimate of the ideal observer filter, the difference between the two unmasked images being discriminated.

Perceptual classification images were estimated for a Vernier discrimination task. The display screen had 48 pixels deg−1 horizontally and vertically. The no-offset image had a dark horizontal line of 4 pixels, a 1 pixel space, and 4 more dark pixels. Classification images were based on 1600 discrimination trials with the line contrast adjusted to keep the error rate near 25%. In the offset image, the second line was one pixel higher. Unlike the ideal observer filter (a horizontal dipole), the observer perceptual classification images are strongly oriented. Fourier transforms of the classification images had a peak amplitude near 1 cycle deg−1 and an orientation near 25 deg. The spatial spread is much more than image blur predicts, and probably indicates the spatial position uncertainty in the task.

Discriminability measures for predicting readability of text on textured backgrounds

Several discriminability measures were examined for their ability to predict reading search times for three levels of text contrast and a range of backgrounds (plain, a periodic texture, and four spatial-frequency-filtered textures created from the periodic texture). Search times indicate that these background variations only affect readability when the text contrast is low, and that spatial frequency content of the background affects readability. These results were not well predicted by the single variables of text contrast (Spearman rank correlation = -0.64) and background RMS contrast (0.08), but a global masking index and a spatial-frequency-selective masking index led to better predictions (-0.84 and -0.81, respectively).

Detection in fixed and random noise in foveal and parafoveal vision explained by template learning

Foveal and parafoveal contrast detection thresholds for Gabor and checkerboard targets were measured in white noise by means of a two-interval forced-choice paradigm. Two white-noise conditions were used: fixed and twin. In the fixed noise condition a single noise sample was presented in both intervals of all the trials. In the twin noise condition the same noise sample was used in the two intervals of a trial, but a new sample was generated for each trial. Fixed noise conditions usually resulted in lower thresholds than twin noise. Template learning models are presented that attribute this advantage of fixed over twin noise either to fixed memory templates' reducing uncertainty by incorporation of the noise or to the introduction, by the learning process itself, of more variability in the twin noise condition. Quantitative predictions of the template learning process show that it contributes to the accelerating nonlinear increase in performance with signal amplitude at low signal-to-noise ratios.

Object detection in natural backgrounds predicted by discrimination performance and models

Many models of visual performance predict image discriminability, the visibility of the difference between a pair of images. We compared the ability of three image discrimination models to predict the detectability of objects embedded in natural backgrounds. The three models were: a multiple channel Cortex transform model with within-channel masking; a single channel contrast sensitivity filter model; and a digital image difference metric. Each model used a Minkowski distance metric (generalized vector magnitude) to summate absolute differences between the background and object plus background images. For each model, this summation was implemented with three different exponents: 2, 4 and infinity. In addition, each combination of model and summation exponent was implemented with and without a simple contrast gain factor. The model outputs were compared to measures of object detectability obtained from 19 observers. Among the models without the contrast gain factor, the multiple channel model with a summation exponent of 4 performed best, predicting the pattern of observer d's with an RMS error of 2.3 dB. The contrast gain factor improved the predictions of all three models for all three exponents. With the factor, the best exponent was 4 for all three models, and their prediction errors were near 1 dB. These results demonstrate that image discrimination models can predict the relative detectability of objects in natural scenes.

Visual signal detection in structured backgrounds. II. Effects of contrast gain control, background variations, and white noise

Studies of visual detection of a signal superimposed on one of two identical backgrounds show performance degradation when the background has high contrast and is similar in spatial frequency and/or orientation to the signal. To account for this finding, models include a contrast gain control mechanism that pools activity across spatial frequency, orientation and space to inhibit (divisively) the response of the receptor sensitive to the signal. In tasks in which the observer has to detect a known signal added to one of M different backgrounds grounds due to added visual noise, the main sources of degradation are the stochastic noise in the image and the suboptimal visual processing. We investigate how these two sources of degradation (contrast gain control and variations in the background) interact in a task in which the signal is embedded in one of M locations in a complex spatially varying background (structured background). We use backgrounds extracted from patient digital medical images. To isolate effects of the fixed deterministic background (the contrast gain control) from the effects of the background variations, we conduct detection experiments with three different background conditions: (1) uniform background, (2) a repeated sample of structured background, and (3) different samples of structured background. Results show that human visual detection degrades from the uniform background condition to the repeated background condition and degrades even further in the different backgrounds condition. These results suggest that both the contrast gain control mechanism and the background random variations degrade human performance in detection of a signal in a complex, spatially varying background. A filter model and added white noise are used to generate estimates of sampling efficiencies, an equivalent internal noise, an equivalent contrast-gain-control-induced noise, and an equivalent noise due to the variations in the structured background.

Image discrimination models predict detection in fixed but not random noise

By means of a two-interval forced-choice procedure, contrast detection thresholds for an aircraft positioned on a simulated airport runway scene were measured with fixed and random white-noise masks. The term fixed noise refers to a constant, or unchanging, noise pattern for each stimulus presentation. The random noise was either the same or different in the two intervals. Contrary to simple image discrimination model predictions, the same random noise condition produced greater masking than the fixed noise. This suggests that observers seem unable to hold a new noisy image for comparison. Also, performance appeared limited by internal process variability rather than by external noise variability, since similar masking was obtained for both random noise types.

Calibration of a Spatiotemporal Discrimination Model from Forward, Simultaneous, and Backward Masking

We have been developing a simplified spatiotemporal discrimination model similar to our simplified spatial model in that masking is assumed to be a function of the local visible contrast energy. The overall spatiotemporal sensitivity of the model is calibrated to predict the detectability of targets on a uniform background. To calibrate the spatiotemporal integration functions that define local visible contrast energy, spatiotemporal masking data are required.

Observer thresholds were measured (2IFC) for the detection of a 12 ms target stimulus in the presence of a 700 ms mask. Targets were 1, 3, or 9 cycles deg−1 sine-wave gratings. Masks were either one of these gratings or two of them combined. The target was presented in 17 temporal positions with respect to the mask, including positions before, during, and after the mask. Peak masking was found near mask onset and offset for 1 and 3 cycles deg−1 targets, while masking effects were more nearly uniform during the mask for the 9 cycles deg−1 target.

As in the purely spatial case, the simplified model cannot predict all the details of masking as a function of masking component spatial frequencies, but overall the prediction errors are small.

Relevant Image Features for Vernier Acuity

The extraction of relevant image features is a key part of discrimination learning (E Gibson, 1969 Principles of Perceptual Learning and Development) and the identification of those features is necessary for the understanding of observer performance. Two mechanisms are thought to limit Vernier acuity judgments: orientation-selective and local-sign mechanisms (Waugh and Levi, 1993 Vision Research33 539 – 552; Beard et al, 1997 Vision Research37 325 – 346). The linear component of relevant image features can be determined for a Vernier task by adding external noise to the image and then averaging the noises separately for the four types of stimulus/response trials.

The Vernier stimulus consisted of two short, dark, horizontal lines presented within low-contrast white noise. Two spatial separations were tested: nearly abutting and a wide horizontal separation. The task was to determine if the target lines were aligned or offset in the vertical direction. The noises were averaged separately for the groups of trials, corresponding to each of the four possible stimulus/response combinations (eg, stimulus=offset, response=aligned). The sum of the two ‘not aligned’ images was then subtracted from the sum of the ‘aligned’ images to obtain an overall image. We then computed a weighted average of adjacent pixels, smoothing the image and fostering visualisation of the relevant feature pattern. Our image analysis supports the view that as the vernier line separation narrows, oriented features play a larger role in the Vernier discrimination.

A hexagonal orthogonal-oriented pyramid as a model of image representation in visual cortex

Retinal ganglion cells represent the visual image with a spatial code, in which each cell conveys information about a small region in the image. In contrast, cells of primary visual cortex employ a hybrid space-frequency code in which each cell conveys information about a region that is local in space, spatial frequency, and orientation. Despite the presumable importance of this transformation, we lack any comprehensive notion of how it occurs. Here we describe a mathematical model for this transformation. The hexagonal orthogonal-oriented quadrature pyramid (HOP) transform, which operates on a hexagonal input lattice, employs basis functions that are orthogonal, self-similar, and localized in space, spatial frequency, orientation, and phase. The basis functions, which are generated from seven basic types through a recursive process, form an image code of the pyramid type. The seven basis functions, six bandpass and one low-pass, occupy a point and a hexagon of six nearest neighbors on a hexagonal sample lattice. The six bandpass basis functions consist of three with even symmetry, and three with odd symmetry. The three even kernels are rotations of 0, 60, and 120 degrees of a common kernel; likewise for the three odd kernels. At the lowest level, the inputs are image samples. At each higher level, the input lattice is provided by the low-pass coefficients computed at the previous level. At each level, the output is subsampled in such a way as to yield a new hexagonal lattice with a spacing square root 7 larger than the previous level, so that the number of coefficients is reduced by a factor of seven at each level. In the biological model, the input lattice is the retinal ganglion cell array. The resulting scheme provides a compact, efficient code of the image and generates receptive fields that resemble those of the primary visual cortex.

CT measures of cerebrospinal fluid volume in alcoholics and normal volunteers

Cranial computed tomography (CT) scans were obtained in 46 male chronic alcoholics and 31 normal male volunteers. Automated methods were used to estimate the cerebrospinal fluid (CSF) volume in various intracranial zones. Measures of the ventricular fluid volume, the volume of fluid in cortical areas on CT sections at the level of the ventricles, and the sulcal fluid volumes on two convexity sections were computed. The alcoholic group, excluding subjects with chronic liver disease, had significantly more fluid than the control group on all sulcal measures. The group difference on the ventricular measure fell short of significance. Within the alcoholic group, no significant correlation was found between the number of years of alcoholism and any fluid measure when normal age effects were taken into account. A striking degree of variability in the sulcal volumes was observed within the alcoholic group, with many subjects showing normal values while a large group showed markedly elevated values. Further studies will be necessary to determine the significance of these variations.

White matter changes in cerebral computed tomography related to aging

Changes on computed cranial tomography related to aging were studied in 123 normal subjects aged 23 to 88 years. The attenuation value of the white matter in the centrum semiovale decreased with age. When the effect of age was excluded, ventricular and sulcal size did not have an independent effect on the attenuation value. This finding suggests that there may be gradual changes in the chemical composition of white matter with aging.

Changes on computed cranial tomography with aging: intracranial fluid volume

A semiautomated computer analysis was developed to estimate fluid volumes in each hemicranium from computed tomography scans. The method was used to estimate total ventricular and sulcal fluid in 123 normal subjects aged 23-88 years. A wide range of normal values was found. The trend was for the estimated ventricular and sulcal fluid volumes to remain relatively constant until age 60 and then to increase at an increasing rate thereafter. Ventricular enlargement occurred in the absence of sulcal enlargement and vice versa. The estimate of the volume of the ventricles was related to skull size. When this was taken into account, the size of the ventricles showed no sex difference. The cranial cavity was larger in men than in women, and, in both genders, the left hemicranium and the left ventricle were larger on the average than their right counterparts. The limitations of computed cranial tomography as a quantitative tool are discussed in detail.