Observing animals and humans: dogs target their gaze to the biological information in natural scenes

Introduction

Visual attention, often measured as ‘looking preferences’, is commonly used to assess animal behavior and cognition (Winters, Dubuc & Higham, 2015). Eye gaze tracking provides more quantifiable and specific information, and during the past decade, it has been successfully applied when investigating dogs’ responses towards still images. Previously, dogs have been found to spontaneously discriminate human and dog faces from each other and from inanimate objects (Racca et al., 2010; Somppi et al., 2012; Gergely et al., 2019) and even facial expressions of humans and dogs (Nagasawa et al., 2011; Somppi et al., 2016; Somppi et al., 2017). Dogs can learn to differentiate facial images of dogs from the facial images of other animal species (Autier-Dérian et al., 2013) and to discriminate between naturalistic images that contain a dog from those that do not (Range et al., 2008). However, most of the studies using looking-time paradigms have used facial images and unlike primate studies (Kano & Tomonaga, 2009; Kano, Call & Tomonaga, 2012) little canine research has used naturalistic whole-body images of humans and animals (Törnqvist et al., 2015), or different kinds of images (e.g., landscapes and faces) embedded in the same study design (Range et al., 2008; Autier-Dérian et al., 2013). Extending the research of canine gazing behavior to more naturalistic images provides a unique opportunity to study both the effect of image composition and object or category content at the same time.

The capacity to respond appropriately to biologically relevant stimuli (prey, predators and social partners) is crucial for animal survival, and it has been suggested to be an unlearned social predisposition (for reviews, Rosa-Salva, Mayer & Vallortigara, 2015; Di Giorgio et al., 2017). The primate visual system is specialized in detecting living creatures from complex natural scenes (Fabre-Thorpe, Richard & Thorpe, 1998; Thorpe et al., 2001), and typically humans and monkeys focus their gaze specifically on the head area of these creatures (e.g., Yarbus, 1967; Kano & Tomonaga, 2009; Kano, Call & Tomonaga, 2012). Biologically relevant stimuli such as animals attract attention more than inanimate objects and social information draws more attention than non-social (e.g., in non-human primates: Machado et al., 2011; Hanazuka et al., 2012; in humans: New, Cosmides & Tooby, 2007; Simion, Regolin & Bulf, 2008; in dogs: Somppi et al., 2012; Törnqvist et al., 2015). Dogs’ spontaneous preferences toward social information has also been recently demonstrated with biological motion videos (Kovács et al., 2016; Ishikawa et al., 2018).

The amount of visual experience a person has in a particular domain can promote an expert cognitive strategy in processing visual information (e.g., Diamond & Carey, 1986; Kundel et al., 2008; Kujala et al., 2012). For example, when viewing images of two dogs interacting with or facing away from one another, dog experts gaze relatively longer at the dogs’ bodies than at their heads when compared with non-experts (Kujala et al., 2012). Dog owners assess the approachability of dogs by assessing their facial expression more rapidly and with fewer fixations than non-dog owners, but when assessing human and monkey faces, both groups have similar viewing behavior (Gavin, Houghton & Guo, 2017). Also human infants, who have previous experience with pets, focus their attention more on the head areas in cat and dog images than infants without pet experience (Kovack-Lesh, McMurray & Oakes, 2014). When humans and non-human primates observe social videos of other animals demonstrating their species-typical behaviors, the distinct viewing patterns exhibited by these observers are influenced by their previous experiences of that species (Kano et al., 2018). Likewise, the amount of previous exposure also affects dogs’ gazing behavior during the presentation of human and dog faces (Somppi et al., 2014; Barber et al., 2016) and natural images of social interaction (Törnqvist et al., 2015).

In the present study, we measured the gazing behavior of family and kennel dogs while they viewed natural images of dogs, humans and wild exotic animals. Three categories of images were used: landscape images containing single humans or animals; full body images containing a human or an animal; and full body images containing a pair of humans or animals of the same species. We hypothesized that when dogs viewed naturalistic images, dogs would focus more attention on living creatures—heads and bodies of animals and humans—even when the figure was small in relation to the image. In addition, we assumed that the composition of the images would affect dogs’ gazing behavior. We also anticipated that dogs would gaze at the heads and bodies of living creatures, even when viewing unfamiliar species of animals; and that dogs would focus more on their conspecifics than non-conspecifics. Furthermore, we assumed that family and kennel dogs’ social experiences would affect the way in which they viewed biological information.

Materials & Methods

Stimuli

Altogether, 36 different stimulus images were presented in this study. The stimuli were color photographs of dogs, humans, and wild animals (mammals, e.g., elephants, tigers, pandas). Three categories of images were used: (1) landscape images containing a human or an animal, (2) full body images of a single human or an animal, (3) full body images of two paired humans or animals (12 images per each category, see examples in Fig. 1). Each image category (landscape, single, paired) contained images of dogs, humans, and wild animals (four images of each subtype in all categories). For example, the landscape image category consisted of images of dogs, humans, and wild animals within a landscape. The physical dimensions of the images were (19.2 × 14.6 cm (height × width). The pixel resolution of the images was 725 × 550 pixels, overlaid on a grey background of 1,680 × 1,050 pixels. Stimuli were obtained from 123RF, Shutterstock, Adobe stock, Wikimedia commons, Microsoft clipart and Bigstock databases.

Results

The total gaze time did not differ between family and kennel dogs (between-subjects factor group, F_1,22 = 0.024, p = 0.877). Instead, main effect of AOI area (F_2,38 = 38.9, p = 0.001) and interaction effects between AOI area × group (F_2,38 = 4.6, p = 0.020), AOI area × species (F_3,62 = 2.9, p = 0.046), AOI area × image category (F_3,69 = 10.1, p = 0.001) and AOI area × image category x group (F_3,69 = 3.2, p = 0.028) were found.

As a main effect, both dog groups gazed at the head area longer than the body (0.10 ± 0.03 and−0.10 ± 0.03, respectively; t₂₃ = 3.3, p = 0.003) or background area (0.10 ±  0.03 and −0.26 ± 0.03, respectively; t₂₃ = 8.6, p = 0.001). Furthermore, the body area was gazed longer than the background area (−0.10 ± 0.03 and −0.26 ± 0.03, respectively; t₂₃ = 3.4, p = 0.002). In addition, the object area was gazed longer than the background area (0.27 ± 0.03 and −0.26 ± 0.03, respectively; t₂₃ = 8.3, p = 0.001; Fig. 2).

Within-groups comparisons showed that family dogs gazed at the head area longer than the body (0.15 ±  0.03 and −0.15 ±  0.03, respectively; t₁₅ = 4.6, p = 0.001) or background (0.15 ±  0.03 and −0.31 ± 0.04, respectively; t₁₅ = 11.0, p = 0.001) areas. In addition, gazing time was longer for the object area than for the background (0.31 ± 0.04 and −0.31 ± 0.04, respectively; t₁₅ = 8.0, p = 0.001) areas in family dogs. Furthermore, the body area gazing time was longer than the background area gazing time (−0.15 ± 0.03 and −0.31 ± 0.04, respectively; t₁₅ = 2.7, p = 0.018). Also in kennel dogs, the gazing time was longer for the head than the background (0.01 ± 0.05 and −0.18 ± 0.04, respectively; t₇ = 3.3, p = 0.012) area. Furthermore, kennel dogs gazed at the object area longer than the background area (0.18 ±  0.04 and −0.18 ±  0.04, t₇ = 4.1, p = 0.005; Fig. 3 & Table 1).

Both dog groups gazed longer at the head area in wild animal images vs. dog images (0.18 ± 0.04 and 0.06 ± 0.06, respectively; t₂₃ = −2.1, p = 0.050, statistical trend) and likewise longer in wild animal vs. human images (0.18 ± 0.04 and 0.07 ±  0.04, respectively; t₂₃ = −2.1, p = 0.043). The body area was gazed longer in images containing dogs vs. wild animals (−0.06 ±  0.06 and −0.18 ±  0.04, respectively; t₂₃ = 2.1, p = 0.050, statistical trend), and also in images containing humans vs. wild animals (−0.07 ± 0.04 and −0.18 ± 0.04, respectively; t₂₃ = 2.1, p = 0.043). In addition, the background was gazed longer in images containing dogs vs. wild animals (−0.22 ± 0.04 and −0.31 ±  0.03, respectively; t₂₃ = − 2.1, p = 0.048; Fig. 4 & Table 2).

Family dogs gazed longer at the head area in images containing a single vs. paired human or animal images (0.22 ± 0.06 and 0.07 ± 0.02, respectively; t₁₅ = 2.7, p = 0.017). Family dogs also gazed longer at the body area in paired vs. single human or animal images (−0.07 ± 0.02 and −0.22 ± 0.06, respectively; t₁₅ = −2.7, p = 0.017; Fig. 5). Furthermore, they gazed longer at objects in single human or animal vs. landscape images (0.49 ±  0.08 and 0.11 ±  0.03, respectively; t₁₅ = −4.9, p = 0.001) and likewise longer in paired human or animal vs. landscape images (0.31 ± 0.06 and 0.11 ±  0.03, respectively; t₁₅ = −3.2, p = 0.007) and also longer in single human or animal vs. paired images (0.49 ±  0.08 and 0.31 ± 0.06, respectively; t₁₅ = 2.4, p = 0.030). Instead, family dogs gazed longer at the background area in landscape vs. single (−0.11 ± 0.03 and −0.49 ± 0.08, respectively; t₁₅ = 4.9, p = 0.001) or paired images (−0.11 ±  0.03 and −0.28 ±  0.05, respectively; t₁₅ = 3.0, p = 0.010). Likewise, they gazed longer at the background areas in paired vs. single images (−0.28 ±  0.05 and −0.49 ±  0.08, respectively; t₁₅ = −3.5, p = 0.003; Fig. 6). For kennel dogs, the gazing times of object area were also longer for paired human or animal vs. landscape images (0.29 ±  0.09 and 0.08 ±  0.06, respectively; t₇ = −4.0, p = 0.005). Kennel dogs also gazed longer at the background area in landscape vs. paired human or animal images (−0.08 ± 0.06 and –0.26 ±  0.07, respectively; t₇ = 3.5, p = 0.009; Fig. 6 & Table 3).

Discussion

In this work, we studied dogs’ visual scanning of images of natural scenes containing dogs, humans, and exotic wild animals. The present work was the first canine eye tracking study using stimuli species other than dogs or humans, and using images, which varied in composition (landscapes and close-up images of humans and animals). Generally, dogs identified the living creatures from the images and focused more attention on their heads relative to the background, which is very much in line with the previous eye-tracking studies (e.g., Téglás et al., 2012; Somppi et al., 2012; Barber et al., 2016). Human and non-human primates are known to concentrate their gaze on the human and animal figures with particular attention paid to the faces of those figures (e.g., Yarbus, 1967; Henderson & Hollingworth, 1999; Kano & Tomonaga, 2009; Machado et al., 2011). Furthermore, human and non-human primates are able to detect living creatures from a landscape image quickly and accurately, even if they are partially hidden and relatively small compared to the background (Fabre-Thorpe, Richard & Thorpe, 1998; Thorpe et al., 2001; New, Cosmides & Tooby, 2007). Our current results suggest a similar, spontaneous detection of living creatures by dogs. The dogs’ preference for biologically relevant information within visual scenes is consistent with the “life detector mechanism”; a basic visual method for focusing attention towards and detecting other animals (reviewed in Rosa-Salva, Mayer & Vallortigara, 2015). Dogs’ preference for biological information is also highlighted in the recent studies using point-light animations of biological motion (Kovács et al., 2016; Ishikawa et al., 2018). In addition to the whole biological creatures, dogs generally fixated more on the heads of animals and humans rather than backgrounds of the images, which further emphasizes the importance of faces in social animals’ visual processing (for a review, Leopold & Rhodes, 2010).

Contrary to our expectations, dogs’ gazing behavior was not greatly altered by the species depicted in the images. Dogs gazed at the head areas in wild animal images longer than in human or dog images. Possibly, the naturally novel wild animals attracted dogs’ attention to a higher degree than humans or dogs (Fantz, 1964), and this kind of natural novelty might reflect the general visual experience of the observer. Previously, rhesus monkeys showed an increased pupillary response, when they viewed videos containing non-primate animals (e.g., marine mammals) and landscapes (e.g., polar), for which they have had limited exposure during their lives (Machado et al., 2011). The ability to detect and respond quickly to novel events is essential for survival in the nature (Ranganath & Rainer, 2003), which is also highlighted in our current results. In previous studies using facial images, dogs preferred to look at conspecific faces over those of humans (e.g., Somppi et al., 2012; Somppi et al., 2014; Bognár, Iotchev & Kubinyi, 2018). However, when they viewed photographs presenting whole bodies within a scene, they have gazed more at humans than dogs (Törnqvist et al., 2015). This difference means that factors other than merely the species influences viewing behavior (Dahl, Logothetis & Hoffman, 2007). For example, human infants show spontaneous gazing preferences toward dog versus human pictures (Gergely et al., 2019), probably owing to the novelty of the stimuli. In our study, no great differences were found between the species depicted, possibly because the images were emotionally neutral full-body figures with scant social gestures and three different types of images were used. In rhesus macaques, gaze distribution in viewing unfamiliar non-conspecific animals varied systematically according to their phylogenetic distance from the viewed species. It was also dependent upon the social and emotional contexts of the image (McFarland et al., 2013).

The present study showed, for the first time, that the image composition affects the dogs’ gazing behavior. In general, dogs spent less time gazing at the object areas in landscape images, where the size of the target was small relative to the image size, which contrasts from the other image types. From this result we infer that dogs may have had difficulty in locating objects within the landscape images. The visual acuity and the ability to detect brightness of the canine visual system is lower than in humans (Pongrácz et al., 2017; for a review, Byosiere et al., 2018). Therefore, the size of stimuli that dogs can see clearly may vary greatly (e.g., review in Byosiere et al., 2018; Tanaka et al., 2000; Lind et al., 2017). In this study, dogs were shown color images, which may also have affected their ability to perceive these images. Dogs have dichromatic color vision, but the data determining which colors dogs can discriminate are controversial (e.g., Neitz, Geist & Jacobs, 1989; Miller & Murphy, 1995; for a review, Byosiere et al., 2018). Recent studies suggest that dogs are unable to distinguish red from green color and that color information might be more important than brightness when dogs are required to choose between stimuli (Kasparson, Badridze & Maximov, 2013; Siniscalchi et al., 2017).

Additionally, family dogs gazed at the head areas of single animals or humans longer than images of paired animals or humans. Conversely, family dogs gazed at the body areas of animals or humans longer in the paired rather than the single image conditions. Thus, family dogs’ gazing preference for the head area was clearly shown in images containing single animals, likely indicating that fixations may have been spread more widely where there were two head areas i.e., in the paired animal images. In these images, body areas may have drawn the attention of family dogs’, because of the social bodily gestures that are present in these images. In the paired images, two animals or humans were standing or sitting close to each other. Interestingly, kennel dogs’ gazing behavior did not differ between head and body areas in the single or paired images. This result is in line with our previous studies, where kennel dogs gazed at scenes of social interaction and facial images less than family dogs (Törnqvist et al., 2015; Somppi et al., 2014). Since family dogs have had more opportunities for social interactions than kennel dogs, they may be more interested and have a greater ability to search whether there are social gestures presented in those images. However, the differences between family and kennel dogs’ gazing behavior may be more evident with socially richer stimuli than the ones used in the current study.

Other than the latter difference between the two study groups, family and kennel dogs gazed at the images in a similar manner, focusing their gaze on the head and object areas of the images. This result suggests that the basic viewing processes are the same regardless of the social living environment and life experiences of the participants. Several previous studies have shown that humans and some other newborn vertebrates have visual predispositions for the head region of living creatures (e.g., Johnson et al., 1991; for a review, Morton & Johnson, 1991). Monkey infants and newly hatched chicks show visual preference for faces, even when there has been no previous exposure to faces (Sugita, 2008; Rosa-Salva et al., 2011). Despite their more restricted social living environment, kennel dogs had daily contact with humans and other beagle dogs. This might have diminished the differences in gazing behavior between family and kennel dogs. In addition, the study populations were not equal in number between family and kennel dogs (2:1), which may have affected our results.

In this study, dogs were trained to remain still during image presentation, because spontaneous movement during the eye tracking recordings can cause serious artifacts and loss of data. However, dogs were not trained to target their gaze at the images or screen. During the experiments, dogs were rewarded with a treat after every image block, which was considered necessary in order to prevent frustration and to ensure that dogs were motivated. It is unlikely that the dogs learnt to associate rewards with certain presented images or that a particular viewing strategy would have been reinforced. Dogs were rewarded only when the screen was blank, and the last image of the block was always different making it hard to learn any specific gazing strategy.

Conclusions

In summary, when viewing photographs of natural scenes, dogs focus their gaze on the informative areas such as the bodies and heads of living creatures, as previously shown for primates. Gazing behavior was dependent on the size and species of the target animals in the images for both dogs living in a socially rich and stimulating family environment and dogs living in a more restricted kennel environment. Nevertheless, there were minor differences in viewing behavior of these dog groups, suggesting that social experiences modulate dogs’ spontaneous viewing patterns when looking at natural scenes.

Supplemental Information