Extending the MNREAD sentence corpus: Computer-generated sentences for measuring visual performance in reading.

The MNREAD chart consists of standardized sentences printed at 19 sizes in 0.1 logMAR steps. There are 95 sentences distributed across the five English versions of the chart. However, there is a demand for a much larger number of sentences: for clinical research requiring repeated measures, and for new vision tests that use multiple trials at each print size. This paper describes a new sentence generator that has produced over nine million sentences that fit the MNREAD constraints, and demonstrates that reading performance with these new sentences is comparable to that obtained with the original MNREAD sentences. We measured reading performance with the original MNREAD sentences, two sets of our new sentences, and sentences with shuffled word order. Reading-speed versus print-size curves were obtained for each sentence set from 14 readers with normal vision at two levels of blur (intended to simulate acuity loss in low vision) and with unblurred text. We found no significant differences between the new and original sentences in reading acuity and critical print size across all levels of blur. Maximum reading speed was 7% slower with the new sentences than with the original sentences. Shuffled sentences yielded slower maximum reading speeds and larger reading acuities than the other sentences. Overall, measures of reading performance with the new sentences are similar to those obtained with the original MNREAD sentences. Our sentence generator substantially expands the reading materials for clinical research on reading vision using the MNREAD test, and opens up new possibilities for measuring how text parameters affect reading.

Spatial-frequency and contrast properties of crowding.

Crowding, the difficulty in recognizing a letter flanked by other letters, has been explained as a lateral masking effect. The purpose of this study was to examine the spatial-frequency and contrast dependencies of crowding, and to compare them with the properties of pattern masking. In experiment 1, we measured contrast thresholds for identifying the middle letters in strings of three randomly chosen lower-case letters (trigrams), for a range of letter spacings. Letters were digitally filtered using a set of bandpass filters, with peak object spatial frequencies ranging from 0.63 to 10 c/letter. Bandwidth of the filters was 1 octave. Frequencies of the target and flanking letters were the same, or differed by up to 2 octaves. Contrast of the flanking letters was fixed at the maximum value. Testing was conducted at the fovea and 5 degrees eccentricity. We found that crowding exhibits spatial-tuning functions like masking, but with generally broader bandwidths than those for masking. The spatial extent of crowding was found to be about 0.5 deg at the fovea and 2 deg at 5 degrees eccentricity, independent of target letter frequency. In experiment 2, we measured the contrast thresholds for identifying the middle target letters in trigrams for a range of flanking letter contrasts at 5 degrees eccentricity. At low flanker contrast, crowding does not show a facilitatory region, unlike pattern masking. At high flanker contrast, threshold rises with contrast with an exponent of 0.13-0.3, lower than corresponding exponents for pattern masking. In experiment 3, we varied the contrast ratio between the flanking letters and the target letters, and found that the magnitude of crowding increases monotonically with contrast ratio. This finding contradicts a prediction based on a grouping explanation for crowding. Our results are consistent with the postulation that crowding and masking may share the same first stage linear filtering process, and perhaps a similar second-stage process, with the additional property that the second-stage process in crowding pools information over a spatial extent that varies with eccentricity.

Psychophysics of reading. XX. Linking letter recognition to reading speed in central and peripheral vision.

Our goal is to link spatial and temporal properties of letter recognition to reading speed for text viewed centrally or in peripheral vision. We propose that the size of the visual span - the number of letters recognizable in a glance - imposes a fundamental limit on reading speed, and that shrinkage of the visual span in peripheral vision accounts for slower peripheral reading. In Experiment 1, we estimated the size of the visual span in the lower visual field by measuring RSVP (rapid serial visual presentation) reading times as a function of word length. The size of the visual span decreased from at least 10 letters in central vision to 1.7 letters at 15 degrees eccentricity, in good agreement with the corresponding reduction of reading speed measured by Chung and coworkers (Chung, S. T. L., Mansfield, J. S., & Legge, G. E. (1998). Psychophysics of reading. XVIII. The effect of print size on reading speed in normal peripheral vision. Vision Research, 38, 2949-2962). In Exp. 2, we measured letter recognition for trigrams (random strings of three letters) as a function of their position on horizontal lines passing through fixation (central vision) or displaced downward into the lower visual field (5, 10 and 20 degrees ). We also varied trigram presentation time. We used these data to construct visual-span profiles of letter accuracy versus letter position. These profiles were used as input to a parameter-free model whose output was RSVP reading speed. A version of this model containing a simple lexical-matching rule accounted for RSVP reading speed in central vision. Failure of this version of the model in peripheral vision indicated that people rely more on lexical inference to support peripheral reading. We conclude that spatiotemporal characteristics of the visual span limit RSVP reading speed in central vision, and that shrinkage of the visual span results in slower reading in peripheral vision.

Invariant recognition of natural objects in the presence of shadows.

Shadows are frequently present when we recognize natural objects, but it is unclear whether they help or hinder recognition. Shadows could improve recognition by providing information about illumination and 3-D surface shape, or impair recognition by introducing spurious contours that are confused with object boundaries. In three experiments, we explored the effect of shadows on recognition of natural objects. The stimuli were digitized photographs of fruits and vegetables displayed with or without shadows. In experiment 1, we evaluated the effects of shadows, color, and image resolution on naming latency and accuracy. Performance was not affected by the presence of shadows, even for gray-scale, blurry images, where shadows are difficult to identify. In experiment 2, we explored recognition of two-tone images of the same objects. In these images, shadow edges are difficult to distinguish from object and surface edges because all edges are defined by a luminance boundary. Shadows impaired performance, but only in the early trials. In experiment 3, we examined whether shadows have a stronger impact when exposure time is limited, allowing little time for processing shadows; no effect of shadows was found. These studies show that recognition of natural objects is highly invariant to the complex luminance patterns caused by shadows.

The effect of contrast on reading speed in dyslexia.

Contrast coding has been reported to differ between dyslexic and normal readers. Dyslexic readers require higher levels of contrast to detect sinewave gratings for certain spatiotemporal conditions, and dyslexic readers show faster visual search at low contrast. We investigated whether these differences in early contrast coding generalize to reading performance by measuring reading speed as a function of text contrast for dyslexic children and adults and for age-matched controls. Contrast affected reading performance of dyslexic and normal readers similarly. For both groups, reading speed was relatively constant between 100 and 2% contrast, and decreased rapidly below 2% contrast. This pattern of results held true for both children and adults, for text with and without sentence context, across a range of character sizes, and for reading aloud and reading silently. We conclude that earlier findings of group differences in contrast effects on grating detection or visual search tasks do not generalize to reading.

Reading with a head-mounted video magnifier.

PURPOSE: This study compared the effectiveness of a head-mounted video magnifier, low-vision enhancement system (LVES), with closed-circuit TV (CCTV) and large print as a device or means of improving reading performance in people with low vision. METHODS: The reading performance of ten low-vision participants was assessed in two ways: (1) By measuring reading speed as a function of print size with LVES and without LVES, and (2) by comparing reading speed and comprehension of news articles using the LVES vs. a popular non-head-mounted video magnifier, the CCTV. RESULTS: Maximum reading speeds with LVES matched the maximum reading speeds with unaided vision attained by enlarging print. The critical print size (the smallest print size that could be read at maximum reading speed) improved significantly for all participants using LVES compared with unaided vision. When comparing reading performance using LVES and CCTV, we found that reading speed and comprehension for the two conditions were equivalent. The two low-vision participants with lowest acuities (20/640 and 20/960) could not read the 10-point newspaper articles with LVES, even with an 8 D auxiliary reading lens that permitted a very close reading distance. CONCLUSIONS: Head-mounted video magnifiers, such as LVES, can support good low-vision reading performance, but the restricted range of magnification may limit the usefulness of the device as a reading magnifier for people with very low acuity.

What is low vision? A re-evaluation of definitions.

PURPOSE: To re-evaluate definitions of low vision, visual impairment, and disability. METHODS: We review current definitions of legal blindness and low vision and how these definitions are variably based on disability or impairment. We argue for a definite distinction being made between criteria for visual impairment and visual disability, low vision being defined as the presence of a visual impairment that results in a disability. Visual impairment is defined according to population norms and a statistical cut-off is used. Visual disability is defined by consideration of the level of visual measures which result in measurable or reportable disability. We consider the evidence that contrast sensitivity should be a criterion for visual disability in addition to visual acuity and visual field. CONCLUSIONS: According to the current information, we define visual impairment as best monocular or binocular visual acuity <(worse than) 6/7.5, total horizontal visual field <146 degrees (Goldmann III-4e) or <109 degrees (III-3e), and contrast sensitivity <1.5 (PelliRobson); we define visual disability as best monocular or binocular visual acuity <6/12 or contrast sensitivity <1.05.

Psychophysics of reading. XVIII. The effect of print size on reading speed in normal peripheral vision.

Reading in peripheral vision is slow and requires large print, posing substantial difficulty for patients with central scotomata. The purpose of this study was to evaluate the effect of print size on reading speed at different eccentricities in normal peripheral vision. We hypothesized that reading speeds should remain invariant with eccentricity, as long as the print is appropriately scaled in size--the scaling hypothesis. The scaling hypothesis predicts that log-log plots of reading speed versus print size exhibit the same shape at all eccentricities, but shift along the print-size axis. Six normal observers read aloud single sentences (approximately 11 words in length) presented on a computer monitor, one word at a time, using rapid serial visual presentation (RSVP). We measured reading speeds (based on RSVP exposure durations yielding 80% correct) for eight print sizes at each of six retinal eccentricities, from 0 (foveal) to 20 deg in the inferior visual field. Consistent with the scaling hypothesis, plots of reading speed versus print size had the same shape at different eccentricities: reading speed increased with print size, up to a critical print size and was then constant at a maximum reading speed for larger print sizes. Also consistent with the scaling hypothesis, the plots shifted horizontally such that average values of the critical print size increased from 0.16 deg (fovea) to 2.22 deg (20 deg peripheral). Inconsistent with the scaling hypothesis, the plots also exhibited vertical shifts so that average values of the maximum reading speed decreased from 807 w.p.m. (fovea) to 135 w.p.m. (20 deg peripheral). Because the maximum reading speed is not invariant with eccentricity even when the print size was scaled, we reject the scaling hypothesis and conclude that print size is not the only factor limiting maximum reading speed in normal peripheral vision.

The viewpoint complexity of an object-recognition task.

There is an ongoing debate about the nature of perceptual representation in human object recognition. Resolution of this debate has been hampered by the lack of a metric for assessing the representational requirements of a recognition task. To recognize a member of a given set of 3-D objects, how much detail must the objects' representations contain in order to achieve a specific accuracy criterion? From the performance of an ideal observer, we derived a quantity called the view complexity (VX) to measure the required granularity of representation. VX is an intrinsic property of the object-recognition task, taking into account both the object ensemble and the type of decision required of an observer. It does not depend on the visual representation or processing used by the observer. VX can be interpreted as the number of randomly selected 2-D images needed to represent the decision boundaries in the image space of a 3-D object-recognition task. A low VX means the task is inherently more viewpoint invariant and a high VX means it is inherently more viewpoint dependent. By measuring the VX of recognition tasks with different object sets, we show that the current confusion about the nature of human perceptual representation is partly due to a failure in distinguishing between human visual processing and the properties of a task and its stimuli. We find general correspondence between the VX of a recognition task and the published human data on viewpoint dependence. Exceptions in this relationship motivated us to propose the view-rate hypothesis: human visual performance is limited by the equivalent number of 2-D image views that can be processed per unit time.

Psychophysics of reading. XVII. Low-vision performance with four types of electronically magnified text.

Most people with low vision need magnification to read. Page navigation is the process of moving a magnifier during reading. Modern electronic technology can provide many alternatives for navigating through text. This study compared reading speeds for four methods of displaying text. The four methods varied in their page-navigation demands. The closed-circuit television (CCTV) and MOUSE methods involved manual navigation. The DRIFT method (horizontally drifting text) involved no manual navigation, but did involve both smooth-pursuit and saccadic eye movements. The rapid serial visual presentation (RSVP) method involved no manual navigation, and relatively few eye movements. There were 7 normal subjects and 12 low-vision subjects (7 with central-field loss, CFL group, and 5 with central fields intact, CFI group). The subjects read 70-word passages at speeds that yielded good comprehension. Taking the CCTV reading speed as a benchmark, neither the normal nor low-vision subjects had significantly different speeds with the MOUSE method. As expected from the reduced navigational demands, normal subjects read faster with the DRIFT method (85% faster) and the RSVP method (169%). The CFI group read significantly faster with DRIFT (43%) and RSVP (38%). The CFL group showed no significant differences in reading speed for the four methods.

Mr. Chips: an ideal-observer model of reading.

The integration of visual, lexical, and oculomotor information is a critical part of reading. Mr. Chips is an ideal-observer model that combines these sources of information optimally to read simple texts in the minimum number of saccades. In the model, the concept of the visual span (the number of letters that can be identified in a single fixation) plays a key, unifying role. The behavior of the model provides a computational framework for reexamining the literature on human reading saccades. Emergent properties of the model, such as regressive saccades and an optimal-viewing position, suggest new interpretations of human behavior. Because Mr. Chip's "retina" can have any (one-dimensional) arrangement of high-resolution regions and scotomas, the model can simulate common visual disorders. Surprising saccade strategies are linked to the pattern of scotomas. For example, Mr. Chips sometimes plans a saccade that places a decisive letter in a scotoma. This article provides the first quantitative model of the effects of scotomas on reading.

Psychophysics of reading--XVI. The visual span in normal and low vision.

The visual span in reading is the number of characters that can be recognized at a glance. The shrinking visual span hypothesis attributes reading deficits in low vision, and slow reading in normal vision at low contrast, to a reduction in the visual span. This hypothesis predicts that reading time (msec/word) becomes increasingly dependent on word length as text contrast decreases. We tested and confirmed this prediction using the rapid serial visual presentation (RSVP) method. Estimates of the visual span ranged from about 10 characters for high-contrast text to less than two characters for low-contrast text. Eye-movement recordings showed that longer reading times at low contrast are partitioned about equally between prolonged fixation times and an increased number of saccades (presumably related to a reduced visual span). RSVP measurements for six out of seven low-vision subjects revealed a strong dependence of reading time on word length, as expected from reduced visual spans.

Motion parallax: effects of blur, contrast, and field size in normal and low vision.

Can people with different forms of low vision use motion parallax to improve depth judgments? We used a staircase method to compare depth thresholds using motion parallax and static viewing. We tested eighteen normal-vision subjects with a range of simulated deficits in acuity, contrast sensitivity, and simulated peripheral-field loss, and ten low-vision subjects with a wide range of acuity, contrast sensitivity, and field loss. Subjects viewed three vertical cylinders monocularly and indicated which one was at a different depth from the other two. For motion-parallax trials, observers moved their heads (in a viewing assembly on rollers) from side to side over a range of 6-12 cm. For static trials, the viewing assembly was fixed in place. Normal-vision subjects' depth thresholds with motion parallax were significantly smaller than those with static viewing by an average factor of 1.95 (p < 0.05) across all levels of acuity and contrast. For low-vision observers, the depth thresholds exhibited large individual differences; however, the motion-parallax thresholds were smaller than the static thresholds by an average factor of 2.05 (p < 0.01). These findings indicate that motion parallax can provide useful depth information for people with low vision.

Psychophysics of reading--XIV. The page navigation problem in using magnifiers.

Most people with low vision require magnification to read. A magnifier's field of view often contains only a few letters at a time. Page navigation is the process by which the reader moves the magnifier from word to word, and from the end of one line to the beginning of the next line. Page navigation takes time and reduces reading speed. The major questions addressed in this paper are: (1) What role does page navigation play in limiting reading speed? and (2) Are the window width requirements for reading (number of characters in the field for a criterion performance level) increased by the need for page navigation? We measured the reading speeds of three normal-vision and seven low-vision subjects in two ways: with drifting-text requiring no page navigation, and with a closed-circuit TV (CCTV) magnifier which required page navigation. We built special hardware to record the location of the CCTV's magnified field in the text. These recordings were used to separate forward-reading time (left-to-right movement through the text) from retrace time (navigational movement). For normal-vision subjects, forward-reading and retrace times were about equal. For low-vision subjects, retrace times were shorter than forward-reading times, indicating that the forward-reading performance was limited by visual, not navigational, demands. The retrace time did have an impact, however, ranging from 17 to 50% of the overall time. The window requirements for reading with page navigation (CCTV) were larger than those for reading without page navigation (drifting-text). The difference was more than a factor of three for normal-vision subjects and close to a factor of two for low-vision subjects (10 characters for CCTV vs 5.2 characters for drifting-text for 85% of maximum reading speed.

Psychophysics of reading. XV: Font effects in normal and low vision.

PURPOSE: Little is known about the effect of font on low-vision reading. In this study, the authors measured the influence of font in reading with normal and low vision. METHODS: Reading acuity, maximum reading speed, and critical print size (the smallest print that can be read with maximum speed) were measured in 50 normal subjects and 42 subjects with low vision. Data were collected using versions of the MNREAD Acuity Chart printed with the Times (proportionally spaced) and Courier (fixed-width) fonts. RESULTS: Reading acuity scores obtained with Courier were better than those obtained with Times for both normal (mean difference, 0.05 logMAR, P < 0.001) and subjects with low vision (0.09 logMAR, P < 0.001). Similarly, critical print sizes measured with Courier were smaller than those measured with Times (mean difference, 0.06 logMAR for normal subjects and subjects with low vision, P < 0.002). Maximum reading speeds for normal subjects were 5% faster with Times than with Courier (P < 0.001), but for subjects with low vision, maximum reading speeds were 10% slower with Times than with Courier (P < 0.05). For print smaller than the critical print size, the reading speeds of normal subjects and subjects with low vision were substantially slower (by as much as 50%) for Times than for Courier. CONCLUSIONS: There are small, but significant, advantages of Courier over Times in reading acuity, critical print size, and reading speed for subjects with low vision. For normal subjects, the differences are slighter, with an advantage in reading speed for Times. However, for print sizes close to the acuity limit, choice of font could make a significant difference in both normal and low-vision reading performance.

The binocular computation of visual direction.

How is a single visual direction assigned to a binocular feature for which the left and right eyes are signaling different directions? According to geometrical principles, binocular visual direction is the average of the visual directions measured from the left and right eyes. Contrary to this prediction, we have found that the relative visual direction between two Gabor targets presented at different stereoscopic depths could be manipulated by varying the contrast ratio between the left and right images. This finding is consistent with a new model in which the relative alignment of depth features is determined from a maximum-likelihood combination of the direction signals from the left and right eyes. In a second experiment we provide support for this model, showing that the magnitude of the contrast-dependent bias in visual direction is predicted by the uncertainty for spatial localization in the left and right images. Lastly we show that visual direction and stereopsis have different dependencies on interocular contrast differences, suggesting that the computation of stereo depth and visual direction are mediated via different mechanisms.

Human efficiency for recognizing and detecting low-pass filtered objects.

Recently, Tjan, Braje, Legge and Kersten [(1995) Vision Research, 35, 3053-3069] found that human efficiency for object recognition was less than 10%, indicating that humans fail to use much of the information available to an ideal observer. We examine two explanations for these low efficiencies: (1) humans are inefficient in using high spatial-frequency information; and (2) humans are inefficient in detecting image samples. We tested the first possibility by measuring human efficiency for recognizing low-pass filtered objects, rendered as line drawings and silhouettes, in luminance noise. Efficiency did not improve when high frequencies were removed, and the first explanation was rejected. We tested the second explanation by comparing efficiencies for object detection and recognition. Recognition efficiency was higher than detection efficiency for silhouettes but not line drawings, showing that detection efficiency does not place a ceiling on recognition efficiency. The results indicate that human vision is designed to extract image features, such as contours, that enhance recognition. A computer simulation suggests that this can occur if the observer views the world through a band-pass spatial-frequency channel.

Human efficiency for recognizing 3-D objects in luminance noise.

The purpose of this study was to establish how efficiently humans use visual information to recognize simple 3-D objects. The stimuli were computer-rendered images of four simple 3-D objects--wedge, cone, cylinder, and pyramid--each rendered from 8 randomly chosen viewing positions as shaded objects, line drawings, or silhouettes. The objects were presented in static, 2-D Gaussian luminance noise. The observer's task was to indicate which of the four objects had been presented. We obtained human contrast thresholds for recognition, and compared these to an ideal observer's thresholds to obtain efficiencies. In two auxiliary experiments, we measured efficiencies for object detection and letter recognition. Our results showed that human object-recognition efficiency is low (3-8%) when compared to efficiencies reported for some other visual-information processing tasks. The low efficiency means that human recognition performance is limited primarily by factors intrinsic to the observer rather than the information content of the stimuli. We found three factors that play a large role in accounting for low object-recognition efficiency: stimulus size, spatial uncertainty, and detection efficiency. Four other factors play a smaller role in limiting object-recognition efficiency: observers' internal noise, stimulus rendering condition, stimulus familiarity, and categorization across views.

Discrimination of compound gratings: spatial-frequency channels or local features?

Models based on spatial-frequency channels and local features provide alternative explanations for suprathreshold pattern discrimination. We compared psychophysical discrimination data with the predictions of the Wilson and Gelb channel model and three local-feature models. The features were peak-valley local contrast, peak-peak local contrast, and luminance gradients. We measured visual sensitivity for discriminating compound gratings (F + 3F or F + 5F, in peaks-add or peaks-subtract phases) whose component contrasts were yoked together so that a contrast increment in one component was accompanied by an equal decrement in the other. The Wilson and Gelb model accounted for the results with peaks-add gratings, but failed to predict those with peaks-subtract gratings. None of the local-feature models explained the results by themselves. Most of the data fell close to an envelope composed of the lowest thresholds of the three feature-detector models, although there were important exceptions. Our findings are consistent with the view that suprathreshold pattern discrimination is mediated by mechanisms responsive to spatially localized features and that more than one type of feature is used.

Psychophysics of reading--XIII. Predictors of magnifier-aided reading speed in low vision.

A key clinical problem in low-vision assessment is finding simple measures that are predictive of real-world performance. Our purpose was to determine whether a set of clinical variables could predict how well low-vision subjects read with their magnifiers. The variables were Snellen acuity, presence/absence of central scotomas, clear/cloudy ocular media, age, magnifier type, and score on a standardized test of reading speed (Minnesota Low-vision Reading Test). Forty low-vision subjects identified the magnifier they most "preferred" to use, based on the frequency and length of time used, ease of use, or any other factor considered important by each subject. All subjects were highly experienced with their magnifier (3 months to 21 yr of use). We classified the magnifiers into five categories: (1) standard correction only; (2) hand-held magnifier; (3) spectacle-mounted magnifier; (4) closed-circuit TV; and (5) stand magnifier. Subjects used their preferred magnifiers to read aloud printed paragraphs of about 150 words from which their reading speeds were calculated. By far the best predictor of the magnifier-aided reading speed was the score on the standardized reading test which accounted for 79.7% of the variance. 43.7% of the variance was accounted for by age, and 42.3% by magnifier type. Snellen acuity, central visual field status, and ocular media status were not significantly correlated with magnifier-aided reading speed. We conclude that a standardized clinical reading test can give a valid prediction of the reading speed a low-vision patient is likely to achieve with a magnifier.