Unresolvable Pixels Contribute to Character Legibility: Another Reason Why High-Resolution Images Appear Clearer

Abstract

This study examined the effect of character sample density on legibility. As the spatial frequency component important for character recognition is said to be 1 to 3 cycles/letter (cpl), six dots in each direction should be sufficient to represent a character; however, some studies have reported that high-density characters are more legible. Considering that these seemingly contradictory findings could be compatible, we analyzed the frequency component of the character stimulus with adjusted sample density and found that the component content of 1 to 3 cpl increased in the high-density character. In the following three psychophysical experiments, high sample density characters tended to have lower contrast thresholds, both for normal and low vision. Furthermore, the contrast threshold with characters of each sample density was predicted from the amplitude of the 1 to 3 cpl component. Thus, while increasing the sample density improves legibility, adding a high frequency is not important in itself. The findings suggest that enhancing the frequency components important for recognizing characters by adding the high-frequency component contributes to making characters more legible.

Keywords: character; critical band; high resolution; legibility; letter; low vision; sample density; spatial frequency.

Figure 1.

Demonstration of characters with different densities, and difference in spatial frequency characteristics with densities. A: The character sample density (sample per letter; spl) was adjusted. 6 spl means a character image using six samples per character height. B: Relationship between the sample density of the images and spatial frequency components of characters. FFT was conducted by imageJ with the character images (a, c, e, o, and s). From the FFT result of each character image, 0° and 90° power (amplitude = sqrt (spectrum)) were averaged. Those with 0 to 13 cpl were plotted for each sample density.

Figure 2.

Example of stimuli used in Experiment 1.

Figure 3.

Average contrast threshold of each sample density. Error bar represents standard error.

Figure 4.

Relationship between predicted contrast threshold and the result of Experiment 1. Error bar represents standard error.

Figure 5.

Relationship between sample density and similarity. The upper panel is a plot of the average stimulus correlation coefficient for each sample density (error bar represents standard error). The lower panel is an example of the similarity of stimuli at sample densities of 6 and 48 spl with contrast 1.0.

Figure 6.

Stimuli used in Experiment 2.

Figure 7.

Relationship between sample density and similarity for stimuli (error bar represents standard error).

Figure 8.

Sample density and spatial frequency components of stimuli in Experiment 2. From the FFT results of each symbol image, 90° power (amplitude = sqrt (spectrum)) of 0 to 13 cpl was averaged for each sample density condition.

Figure 9.

Average contrast threshold of each sample density. Error bar represents standard error.

Figure 10.

Relationship between predicted contrast threshold and the result of Experiment 2. Error bar represents standard error.

Figure 11.

Average contrast threshold of each sample density. Error bar represents standard error.

Figure 12.

The result of Experiment 3 was replotted for each participant’s visual acuity. The horizontal axis represents the retinal sample density of characters (samples/deg).

Figure 13.

The result of Experiment 3 was replotted for each participant’s visual acuity. The horizontal axis represents the spatial sample density of characters (samples/letter-height).

Figure 14.

Relationship between predicted contrast threshold and the result of Experiment 3. Error bar represents standard error.

Figure 15.

Results of Experiments 1 to 3. Standardized at 48 spl for each individual.