Our current environment is characterized by frequent cues for highly palatable foods. Many researchers partially attribute rising obesity rates and problems in eating-related self-regulation to this factor (Meule and Vögele, 2013). To examine the factors underlying appetitive responses to foods, research is increasingly using food images (Van Der Laan et al., 2011). Visual food cues constitute, like odors, a primary sensory input that allows predictions about the edibility and palatability of a food object. Thus, visual food cues can be regarded conditioned stimuli that are associated with the hedonic and homeostatic effects of ingestion and are therefore themselves rewarding (Dagher, 2012). Also, overt eating behaviors are under strong conscious control and therefore do not always reveal underlying response tendencies. Using food images, neurocognitive and indirect measures have been particularly successful in the study of subtle appetitive and regulatory determinants of overt eating behavior.

The “picture viewing approach” is validated by several lines of evidence. First, food deprivation/hunger affects the response to food images on several levels. Short term food deprivation affects responses to food pictures as demonstrated for implicit food evaluation (e.g., Seibt et al., 2007; Hoefling and Strack, 2008), salivation (e.g., Wooley and Wooley, 1981), autonomic responding (e.g., Rodriguez et al., 2005), visual attentional processing (e.g., Stockburger et al., 2009b) and neural reward system activity (e.g., Labar et al., 2001; Uher et al., 2006; Castellanos et al., 2009; Goldstone et al., 2009). Second, food image responses reliably differentiate individuals with abnormal eating behavior from healthy controls: altered food cue processing has been reported in individuals with restrained (Blechert et al., 2010; Burger and Stice, 2011), external (Nijs et al., 2009) or emotional eating (Bohon et al., 2009), as well as in patients with eating disorders (Blechert et al., 2011; Nikendei et al., 2012) or obesity (Nijs and Franken, 2012; Martens et al., 2013). Third, food picture viewing tasks have been adapted to train eating control, for example, through food-specific inhibition tasks (e.g., stop-signal task; Van Koningsbruggen et al., 2013) or attentional retraining (Werthmann et al., 2013; Kakoschke et al., 2014; Kemps et al., 2014) with measurable effects on actual food intake, supporting the notion that responding to food images is causally involved in eating behavior. In sum, there is good evidence that the food picture viewing approach is a useful tool for the study of eating behavior and appetitive/motivational brain systems. However, a number of factors need to be taken into account during selection of images to be able to draw firm conclusions. These factors broadly fall into the categories (1) food types, (2) individual differences, and (3) image characteristics.

Regarding food types, cultures around the world have brought about a vast variety of foods that researchers need to consider when designing experiments. What might be the dimensions that need to be considered during image selection? First, foods differ in caloric content, which has been shown to affect early electrocortical responses (Toepel et al., 2009; Meule et al., 2013). However, caloric density often goes along with the degree of food processing: processed foods are often more energy-dense than natural, unprocessed foods. Processed foods furthermore differ in their colors from whole foods like fruits and vegetables, which utilize the entire color spectrum. Thus, image selection according to caloric density should simultaneously consider level of processing and colors. Besides caloric density, macronutrients, that are proteins, fats, and carbohydrates, should be taken into account, if craving for certain types of food is a construct of interest (e.g., craving for carbohydrates, Corsica and Spring, 2008). Furthermore, there are distinguishable food classes such as vegetables, meat-containing dishes, fruits, and snacks which each differ in their (seasonal) availability, readiness to eat, flavor, nutritional composition, healthiness, color, and familiarity. Obviously, the categorization of foods into some classes is dependent not only on individual experiences and availability of certain foods but also on the research questions asked. It is for that reason, that food-pics provides a variety of food images that cover many food classes and that, most importantly, can be classified as needed by the user.

Not only is there a wide variety of food types to choose from but researchers need to consider the targeted population and therefore individual differences for image selection. For example, if vegetarians or vegans are part of the sample, meat containing images should probably be avoided as these trigger altered neural and behavioral responses in vegetarians compared to omnivores (Stockburger et al., 2009a). Similar considerations apply to food preferences based on cultural, religious or health grounds (Hoffman et al., 2013). Individual preferences affect brain responses, which is why some studies individualize stimuli to match each participant's preferences (e.g., Hollmann et al., 2012; Giuliani et al., 2013). Further individual differences in age and gender, educational status, and body mass index (BMI) should be considered for images selection (Caine-Bish and Scheule, 2009; Raffensperger et al., 2010; Berthoud and Zheng, 2012).

A third class of factors are image characteristics. Unfortunately, dimensions such as brightness, contrast, or spatial frequencies have not received much attention in studies using food-related images. However, effects of such image features on visual perception and stimulus-evoked neuronal responses are well known. Consequently, it is recommended to carefully control the physical properties of visual stimulus material (Knebel et al., 2008; Willenbockel et al., 2010; Kovalenko et al., 2012; Ball et al., 2013). For example, the role of image complexity and spatial frequencies for neural responses are heavily debated in the field of face processing (Vuilleumier et al., 2003; Thierry et al., 2007; Rossion and Jacques, 2008) and are increasingly considered during affective picture viewing (Bradley et al., 2007; Delplanque et al., 2007; Wiens et al., 2011). Thus, similar standards must apply to studies using food pictures rendering standardized stimulus sets and associated meta-data essential. To this end, Foroni et al. (2013) recently presented an image database featuring food (natural, transformed), rotten food, non-food objects (natural, artificial), animals, and scenes along with normative ratings by 73 healthy participants and physical image properties (size, brightness and spatial frequency). They focused on the natural (food, non-food) and artificial distinction in their data analyses. While their database (Foodcast Research Image Database, FRIDa) represents an important step forward in the field of food picture research and their variety of images is broad, the number of edible food items is relatively restricted and, in part, specific to the Mediterranean cuisine. Moreover, their normative data stem from a small sample with little demographic diversity, resulting in a relatively low number of ratings per image.

In the present study, we present food-pics, a stimulus set of 568 food and 315 non-food images. In our study design (normative ratings, image characteristics), we aimed to be complementary to FRIDa and at the same time address some of its limitations. Food-pics was aggregated to represent a wider range of foods to allow applicability in western countries. Our normative participant samples (N = 1988) were selected to represent typical university student samples but, in total, span a comparatively wide range of age (11–77 years), BMI (12–67 kg/m²), and cultural background (German-speaking countries and the USA), to provide robust and generalizable normative data on commonly used perceptual and psychological parameters like palatability, desire to eat, recognizability, familiarity, valence and arousal. Physical image characteristics, that is, color, size, contrast, brightness, and complexity, were computed to complement the dataset and allow the selection of physically matching groups of images. Our analyses explore several example dimensions relevant to study design: (1) image type (e.g., food vs. non-food images) and food type (e.g., vegetables vs. meat vs. fruits, high- vs. low-calorie dense food, sweet vs. savory food, whole vs. processed food), (2) individual differences (e.g., demographics such as age, gender, and BMI, but also cultural background and vegetarianism) and (3) state variables (e.g., hunger and current weight reduction diet) on image ratings. We also explored the relationship of (4) image characteristics (e.g., contrast, brightness, complexity) with subjective ratings and nutritional content.

The present study presents food-pics, a database of images of foods for experimental research on food perception and eating behavior. Previous studies are limited considerably in stimulus selection and/or characterization of stimulus material and food contents hampering the comparability of findings across laboratories. Food-pics comprises a large variety of foods and non-foods along with detailed data on image characteristics, food contents, and normative ratings. We presented example analyses of food types, individual differences, state effects, and image characteristics to explore key variables relevant for experimental design of food viewing studies.

Regarding food types, our results confirm that calorie content is a relevant determinant of subjective responses, in line with a several studies showing distinct neural responses for high- vs. low-calorie images (e.g., Killgore et al., 2003; Cornier et al., 2007; Toepel et al., 2009; Frank et al., 2010). Interestingly, our normative data suggest slightly lower palatability and desire to eat ratings for high-calorie images (small to medium effect size), possibly reflecting the rising awareness of the unhealthy nature of these foods in the populations studied here or self-presentation biases. Other self-report studies show the opposite (Richter et al., under review), as do implicit measures (Houben et al., 2012). It is possible that food restrictions prior testing played a role here because food deprivation renders particularly high-calorie foods more attractive (Goldstone et al., 2009). Our data indicate that participants were not very hungry [M = 28.5, SD = 25.4, on a 1 (not hungry) to 100 (very hungry) scale] but hunger correlated slightly stronger with palatability/desire to eat ratings of high- compared to low-calorie images. Sweet compared to savory foods were rated more palatable and with stronger desire to eat, as were whole vs. processed foods. One has to keep in mind that we used all images of the respective type of the database so it might well be that certain subcategories with a high number of images contributed more than others (e.g., 76 images displayed fruit in the whole and sweet categories). Together these results suggest that image selection will substantially influence (rating) results, depending on the proportion of high-calorie, sweet and whole foods in a specific category. Processed foods are often higher in caloric density, however, researchers could still match the total amount of calories displayed in the images between whole and processed foods by selecting pictures with larger amounts of whole foods (e.g., wild berry mix, 53,75 kcal, image #214) and pictures with smaller amounts of processed foods (e.g., 4 pretzels, 44 kcal, #494). Although recognizability and familiarity of the objects were relatively high, it should be noted that participants performed a yes/no task and did not name the objects.

Individual differences such as restraint, external or emotional eating, eating disorders, or obesity are central independent variables in the study of eating behaviors. However, sampling error can induce group differences on other individual difference variables unless carefully stratified. Age and BMI differences are tolerable to some degree because they showed only minor influence on ratings in our analyses (rs < 0.117). Gender and vegetarianism are more relevant for sampling/matching because lower ratings for palatability and desire to eat were found for women in general and for vegetarians specifically for meat-containing foods. These results reflect in part also inconsistencies in the literature with regard to gender: women are sometimes reported to experience cravings more frequently (Cepeda-Benito et al., 2003) but also restrain and worry about their eating more than men (Dinkel et al., 2005). The present data suggest that in a large unselected sample and across a wide range of foods, women give lower palatability/desire to eat ratings. Thus, normative ratings provided along with the images are reported separately for vegetarians and omnivores and for males and females to facilitate selection of suitable images.

State variables like hunger are obviously important in the food context. Hunger influenced desire to eat to a higher degree than palatability, which is in line with findings that specific state cravings correlate with food deprivation (Cepeda-Benito et al., 2003; Meule et al., 2012) and interesting in the context of the discussion whether “wanting” (~desire to eat) and “liking” (~palatability) are dissociable in humans (Finlayson et al., 2007; Havermans et al., 2009; Finlayson and Dalton, 2011; Havermans, 2011). Hunger might further interact with caloric density as discussed above. Interestingly, current dieting did not influence results much: only a small increase in desire to eat was found for dieters as compared to non-dieters. The literature on dieting effects is mixed: some studies have found dieting to decrease food cravings (reviewed in Martin et al., 2011) other studies found the opposite (Massey and Hill, 2012). On the other hand, weight reduction has been reported to lead to long term weight gain (so called “yoyo effects”) although the mechanisms are not clear (Ochner et al., 2013). These findings underscore the necessity to assess the short and long-term dieting status, the diet success (i.e., weight reductions) as well as the current hunger levels of the participants.

Just as the appearance of foods influences their acceptability for consumption (Wadhera and Capaldi-Phillips, 2014), image characteristics have been shown to affect neurocognitive measures (Vuilleumier et al., 2003; Bradley et al., 2007; Delplanque et al., 2007; Thierry et al., 2007; Rossion and Jacques, 2008) which is why we calculated the key measures that have been established in the field of visual perception (Knebel et al., 2008; Willenbockel et al., 2010; Wiens et al., 2011; Kovalenko et al., 2012; Foroni et al., 2013). Lacking neurocognitive measures in the present study, we explored their relationship with normative ratings and macronutrients. Red color went along with higher arousal ratings whereas green color was indicative of lower calories and lower concentrations of protein, fat, and carbohydrates. Colors should therefore be considered in the study design. Expectedly, our objective index of complexity (reflecting the number of object-components displayed in the image) correlated positively with rated complexity. However, the low to medium sized correlation indicates that subjective and objective measures of complexity are partially independent constructs and studies need to make their pick of which index to use depending on study aims. Future studies might further measure image aesthetics which was not measured here but might be related to expected palatability. Further research should also employ neurocognitive measures to determine which objective and subjective image characteristics influence neural responses. In the lack of such evidence, researchers could use food-pics metadata to match image sets on factors unrelated to their independent variable, particularly when comparing different food types against each other. For example, if the influence of caloric density is to be examined, high- and low-caloric density image sets could be matched for total amount of calories in the image, sweet/savory and processed/whole food proportion, and green color contribution to increase the specificity of the comparisons. If matching is not possible or not desired, researchers should still describe their images in more detail using the metadata provided with food-pics or list the image numbers in a footnote or supplementary material.

In conclusion, we hope that food-pics will facilitate experimental research on food perception, eating behavior and appetitive responses. Databases such as food-pics will increase the comparability of study results and therefore facilitate research communication as it is the case in object recognition, face processing or emotional picture viewing. Food-pics as well as normative rating data can be downloaded free of charge from the first author's website at www.food-pics.sbg.ac.at upon completion of appropriate license agreements.