In the experiment, we randomly chose 20 images for each block, from the original set of 160 images. After giving subjects a sheet with instructions, they performed a training block to ensure that they understood the task and to familiarize themselves with the filtered stimuli. This training situation consisted of a block of 20 trials which included each of the different types of stimuli to be presented. Although it was a training situation, noise was presented prior to the introduction of the filtered image.

After the training block, subjects were presented with 8 blocks of 20 trials of images, each filtered in different spatial frequency. For each trial (see Figure 1), the fixation point was presented in the center of the computer screen for 150 milliseconds to fix the subject’s attention. Next, the filtered noise was presented for 500 milliseconds. Finally, the filtered image was presented for 10.000 milliseconds. Visual noise and image where presented in the same band-pass filter.

Figure 1 Figure 1 Example of the trial used Source: own work

The experiment was conducted in a dimly-lit room, isolated from external noise. Each subject was seated in an armchair in front of a TFT 21 inch computer screen that showed visual stimuli. The screen had a resolution of 1024x768 pixels. Subjects were seated 45 cm from the screen with a chinrest to help them maintain a constant viewing distance. At this distance, the stimuli subtended 7.64 × 7.64 deg and were presented at a rate of 50 Hz. Subjects answered by pressing a keypad button and giving the name of the image presented. RTs were recorded, automatically. The response screen remained either until the subject answered by pressing the keypad button or for a maximum of 10000 ms. The entire experiment lasted approximately 10 minutes.

Data analysis was performed using the statistical program, Statistical Product and Service Solutions (SPSS) v. 13 for windows. We used a linear, mixed design (A₂ x B₃ x C₄) with repeated measures for three factors. Our design includes three independent variables and two dependent variables, each one with different levels.

Factor A represents the independent variable amount of detail, with two levels: greater amount of fine detail (lamp, key, glasses, bicycle, and alarm clock) and greater amount of broad detail (shoe, hat, coffeepot, telephone, and car). Factor B represents the independent variable size, with three levels: small (key, glasses, and alarm clock), medium (shoe, hat, telephone, lamp, and coffeepot), and large (car and bicycle). Factor C represents the independent variable frequency bands, with four levels: level 1 (V_2-8), level 2 (V_10-16), level 3 (V_18-24), and level 4 (V_26-32).

The two dependent variables that were controlled were RT and failures/successes (based on recognition versus non-recognition of images).

We also controlled the maximum presentation time of the image (10,000 ms), the sequence number of the presentation, the luminance of images, the spatial frequency with which images were presented, and the visual noise.

All experimental conditions (ambient noise, room lighting, and isolation of external stimuli) were identical for all subjects.

The results show that the optimum RT for identifying an image filtered in different spatial frequency bands was approximately 2000 ms of exposure. Specifically, familiar and medium-sized stimuli presented in banks of spatial frequency bands with Gaussian filters of variance V26-32 were recognized in a mean time of 2126 ms.

The data also showed that RTs to images composed of fine detail were significantly shorter than those composed of broader detail (F = 69.13; p < 0.0005). Reaction times were also shorter when the stimuli were of medium size (F = 92.24; p < 0.0005). In regards to the established banks of spatial frequency bands, paired comparisons revealed significant differences (F = 45.88; p < 0.0005) for level 4 (V_26-32) versus the other bands analyzed. The comparisons for level 3 (V_18-24) and level 2 (V_10-16) also presented significant differences (p < 0.0005) versus the other frequency bands, but did not register significant differences when compared to each other (p = 0.745). Similar findings were obtained in the comparison of level 2 (V_10-16) and level 1 (V_2-8) (p = 0.067). The fastest responses were reported when medium-sized images were viewed, although there were no significant differences in regards to the amount of detail those responses contained (p = 0.256). Reaction times were longer for large stimuli; here, there were significant differences according to the amount of detail those responses contained (p < 0.0005) (see Figure 2).

Figure 2 Figure 2 Mean values Reaction Time: Detail x Size Source: own work

In the entire size grouping, as the images were presented in higher spatial frequency bands, the speed of recognition showed a progressive increase, although this progression was slightly higher in the group of medium-sized stimuli (see Figure 3).

Figure 3 Figure 3 Mean values Reaction Time: Frequency Band (4 groups) x Detail Source: own work

It should be noted that the real size of the object represented by each image and the time taken to recognize each stimulus were proportionally related to each other. In regards to the amount of detail, the fastest responses were produced by images presented with fine detail and high spatial frequency bands (level 4 with Gaussian filters between V₂₆ and V₃₂). Furthermore, the observed differences were significant (p < 0.0005), with the exception of level 2 spatial frequency bands (V_10-16) (see Figure 4). The research also found that, although the RTs to images incorporating fine detail were shorter than those which incorporated broader detail, the interaction of detail and frequency suggested that this was only the case where high spatial frequencies were also a factor. With low spatial frequencies, regardless of the level of detail (fine vs. broad), there was no difference in the RTs.

Figure 4 Figure 4 Marginal Means: Frequency Band (4 groups) x Size Source: own work

In the present study, we have begun considering the information that spatial frequencies provide in the recognition of faces, objects, and scenes. Specifically, we have taken into account the studies of Morrison and Schyns (2001), Oliva and Schyns (1997), and Torralba and Oliva (2003). In these studies, a single static stimulus is taken into account, while we have considered stimulus sequences. Progressively filtered images were presented to explore how spatial information provided in serial fashion is able to guide the initial stages of early visual processing. In these presentations, we have controlled luminance, contrast, display time, refresh rate of the monitor, displayed image distance, visual angle, and visual noise.

Our results show that the RT in which a subject can determine whether a filtered image at a different spatial frequency band is recognized is located below 10 000 ms of exposure. In fact, we can consider that the RT would be approximately 2 000 ms. Specifically, familiar and medium-sized stimuli presented in banks of spatial frequency bands with Gaussian filters of variance between V26-32 were recognized in a mean time of 2126 ms. Shorter stimulus duration may be one of the possible factors causing low efficiency in discrimination tasks (Gold et al., 1999; Thurman & Grossman, 2011).

In relation to the independent variable prevalence of detail, the results of this research have shown that the RTs are lower when incorporating fine detail conditions than they are when including broad detail conditions. This finding is consistent with the fact that the fastest responses also occur in images with predominantly fine detail in high spatial frequency bands because fine details are perceived in the higher frequency bands, whereas broad details are identified in low frequency bands. With low spatial frequencies, the RT under broad and fine detail conditions is not significant. This finding leads us to believe that the decrease of RTs is due to the presentation of fine detail images in high spatial frequency bands. The progressive sequence of filtering implies that the representation in a space-image scale is an ordered hierarchy ranging from the lowest frequency scales to the highest. As we increase the Gaussian variance (σ²), the fine details of the image (maximum responses of the filters, edges, high spatial frequencies, or fine scales) will appear.

In reference to the size variable, we have found that the fastest responses take place in medium-sized images without significant differences in the characteristics of detail. This does not happen with small and large images where RTs are longer and significant differences do exist between characteristics of detail. Average time for correct answers is lower for medium-sized images than it is for small or large images. In fact, in all responses grouped by size category, recognition of objects is higher as the spatial frequency bands increase. This may be due, as noted previously, to the contrast sensitivity assessment which takes into account the size of the object perceived because the contrast sensitivity varies depending on whether the image is small (i.e. glasses), medium (i.e. coffeepot), or large (i.e. car). We also believe that this phenomenon may be attributed to a relationship of proportionality. When the images are presented to subjects, they expect them to be close to their real size. Our results suggest that the real size of the object represented in each image is related to recognition time.

In relation to the third variable, the bands of spatial frequency, we find faster response times in the higher bands (bank bands nº. 4 with Gaussian filters between V26 and V32). Significant differences in all banks of bands ranging from small to large sizes, coupled with the RTs in the condition of large and small sizes, differ significantly in bank bands nº. 2 and nº. 4. This finding leads us to consider that the decrease of RT is due to the presentation of stimuli predominantly incorporating fine detail in high spatial frequency bands (Díaz Pardo, Suárez Fajardo, Puerto-Leguizamón, & Zona Ortiz, 2015). The spatial frequency bands used in this research are between a Gaussian variance of V2 to V32 (grouped into four banks). What we observe is not different from other experiments on Gaussian filters. Campbell and Robson (1968) show that the detection threshold of complex sinusoidal gratings with high spatial frequencies coincided with the threshold obtained for the fundamental harmonic component. Finally, it should be noted that in relation to the bands, subjects demonstrate a tendency towards faster learning in regards to the stimuli presented because when a stimuli is recognized in a low frequency band and answered correctly, the RT decreases significantly when the same stimuli is presented immediately but in a higher frequency band.