Images of objects are essential materials in many fields of psychological research, such as visual perception, language, memory, and attention (Schiano and Watkins, 1981; Bonin et al., 2002; Goodale and Westwood, 2004; Filliter et al., 2005; Gomez and Shutter, 2008; Denkinger and Koutstaal, 2014). Studies of neural mechanisms and clinical research also use pictures to explore cognitive processes (Hirsh and Funnell, 1995; James et al., 2002; Hawco et al., 2013; van de Nieuwenhuijzen et al., 2013). However, some variables of these images are differ substantially, which can affect object recognition and naming. For instance, objects with high name agreement are named more quickly than low-name-agreement objects, because the lower the name agreement, the more names there are for one object (Vitkovitch and Tyrrell, 1995). A growing number of studies have investigated the contributions of the age of acquisition (AOA) of the object concept to word and picture identification (Juhasz, 2005). Furthermore, viewpoint agreement has been found to affect object identification and object recognition memory (Bülthoff and Newell, 2006; Gomez and Shutter, 2008). Larger deviations between the object viewpoint and the canonical viewpoint correspond to lower efficiency of object perception. In addition to name agreement, AOA, and viewpoint agreement, other variables that influence cognitive processes include visual complexity and concept familiarity. Researchers who wish to use these images in their studies should control for irrelevant variables of stimuli or systematically balance these variables across experimental conditions to facilitate an accurate and reasonable interpretation of the results. Therefore, to decrease the influence of image features and to explain experimental results appropriately, it is essential to normalize object images.

Previous research has primarily employed three types of standardized pictures: black-and-white line drawings, colorized versions of these line drawings, and color images (photographs). Black-and-white line drawings consist of hand drawings based on selected concepts. A pioneering normalization study of line drawings was conducted by Snodgrass and Vanderwart (1980). They presented a set of 260 black-and-white line drawings in English that were standardized on four key variables: name agreement, image agreement, familiarity, and visual complexity. Subsequently, several similar line-drawing normalization efforts have been pursued in different languages (Barry et al., 1997, for British; Bonin et al., 2003, for French; Cuetos et al., 1999, for Spanish; Liu et al., 2011, for Mandarin Chinese; Nishimoto et al., 2005, for Japanese; Pind et al., 2000, for Icelandic; Sirois et al., 2006, for Canadian French). These standardized sets of object images have been widely used in many research fields; they have even been used to compare the influence of different cultures on cognitive processes (Bates et al., 2003; Yoon et al., 2004; Łuniewska et al., 2016).

However, certain pictorial features, such as color or texture, are reduced in typical line drawings. Earlier studies have found that color is a useful cue for object processing (Price and Humphreys, 1989; Wurm et al., 1993; Tanaka and Presnell, 1999). Consistent with previous studies, Rossion and Pourtois (2004) found that adding color information could improve name agreement. Uttl et al. (2006) additionally found that objects with color were easier to identify than black-and-white photos. Therefore, to meet the need for color images in experiments, researchers added color and texture information to the original Snodgrass and Vanderwart (1980) pictures to produce norms for colorized versions of line drawings (Rossion and Pourtois, 2004). Subsequently, these colorized pictures were normalized by other researchers in different languages (Tsaparina et al., 2011; Bakhtiar et al., 2012; Bonin et al., 2013; Raman et al., 2014).

Although color and texture information make black-and-white line drawings much more similar to real-life objects, these pictures still lack ecological value. In particular, they cannot satisfy the requirements of experiments that aim to study real-life situations (Brodeur et al., 2010). For this reason, several investigators have considered collecting sets of color images (photos). In the early phase, researchers provided a small set of photographs of objects, which were downloaded from the web or other online sources (Viggiano et al., 2004; Adlington et al., 2009). In recent years, an increasing number of large sets of photo materials have been standardized in different languages (Brodeur et al., 2010, 2014; Moreno-Martínez and Montoro, 2012; Shao and Stiegert, 2016).

However, there are few normalized datasets of color images in Mandarin Chinese; most normalized images still consist of black-and white line drawings (Shu et al., 1989; Zhang and Yang, 2003; Liu et al., 2011). To our knowledge, only one study has reported normative data for colorized pictures in Chinese (Weekes et al., 2007). That study also suggested that color was an integral part of the object representation. However, the images used in that work were the colorized drawings of Snodgrass and Vanderwart (1980) produced by Rossion and Pourtois (2004), not real color images of objects. Thus, the problem of ecological validity may still exist.

Accordingly, the present work aimed to collect a large sample of color images of common objects and to normalize these images according to several important variables: name agreement, concept familiarity, visual complexity, AOA based on adult ratings, object agreement, and viewpoint agreement. These features were standardized due to their potential effects on cognitive processes.

Table 1 provides norms for the 435 color images on each variable. For the naming task, we included the naming latency as the dependent variable, word frequency and the two measures of name agreement: H-value and percentage name agreement. In addition, concept agreement was calculated according to the proportion of participants who provided the correct concept, the word length of the dominant name was calculated in numbers of characters, and the target name agreement was calculated based on the concepts from Liu et al. (2011). The results of each rating task are listed below. Skewness values for the features are provided to show their distribution. All the standardized variables for each item are provided in the Supplementary Materials (Data Sheet 1).

As in previous normalization studies, after computing descriptive statistics, we performed a correlation analysis to examine how the norms were related to each other. Correlations between the norms are presented in Table 2. The data showed that all variables except visual complexity were related to naming latencies. Visual complexity was significantly related to only familiarity (r = −0.348, p < 0.01) and word frequency (r = 0.146, p < 0.01). The two measures of name agreement were highly correlated (r = −0.912, p < 0.01), and the two measures of image agreement were significantly correlated (r = 0.306, p < 0.01).

The correlation analysis indicated that most variables had highly significant correlations with each other (see Table 2). We then performed factor analysis to explore the internal structure of ten correlated variables; the results are shown in Table 3. The KMO was.661, and Bartlett's test of sphericity produced a result of χ²(45) = 1967.131, p < 0.001, indicating that the factor analysis were effective for the correlated variables. Four factors were extracted from the ten variables. The first factor, name agreement, comprises the percentage name agreement, H-value, and concept agreement. Prior studies have found that the effect of AOA may occur at the lemma level (Belke et al., 2005). In the present study, we found that AOA, word frequency, and word length loaded highly on the second factor, lexicon. Familiarity and visual complexity loaded highly on only the third factor, difficulty of object processing, since visual complexity probably reflects image recognition (Weekes et al., 2007), and several studies have shown that familiarity correlates significantly with visual complexity (Shu et al., 1989; Zhang and Yang, 2003; Weekes et al., 2007; Liu et al., 2011). The last factor, image agreement, consists of object agreement and viewpoint agreement (Brodeur et al., 2010). Altogether, these components accounted for 74.86% of the total variance.

For the 10 important variables, we first used simultaneous multiple regression analysis to identify variables with significant effects on naming latency. However, we found that the percentage name agreement showed a multi-collinearity problem (variance inflation factor, VIF, was greater than 10; tolerance value was 0.094). When the largest VIF (the reverse of tolerance) is in excess of 10, there may be multi-collinearity in the regression model (O'Brien, 2007). We then used stepwise multiple regression analysis to identify the predicting effect of variables on naming latency; the results are shown in Table 4. The VIF and tolerance values suggested that multi-collinearity had little effect on the regression model. The regression analysis yielded an adjusted R² = 0.547, F_{(6, 417)} = 86.290, and p < 0.001. Name agreement (%), familiarity, word frequency, concept agreement, AOA and object agreement showed significant contributions to naming latencies. No other variables were significant.

On the basis of the naming latencies, pictures were divided into five difficulty levels with 87 items in each level. The mean value of all the variables in each level is shown in Table 5. In addition to visual complexity, most variables showed clear increasing or decreasing trends accompanying increases in naming latencies. These measures will provide substantial additional information for future studies of cognitive processes.

This study is the first to provide norms for a large set of color images in Mandarin Chinese, and it standardized several important variables: name agreement (H-value and the percentage of dominant name), concept agreement, word frequency, word length, AOA, familiarity, image agreement (object agreement and viewpoint agreement), and visual complexity. We examined correlations between the norms and explored the internal structure among these correlative variables by a factor analysis. Four factors were extracted and accounted for a large portion of the total variance (74.86%). Furthermore, we used multiple regression analysis to identify variables with significant contributions to naming latencies, including norms of name agreement, familiarity, word frequency, concept agreement, AOA, and object agreement. These variables explained 54.70% of the total variance of naming latencies.

Our main purpose was to establish Mandarin Chinese norms for photo stimuli. Since most of the operations in the current study were conducted with reference to previous studies, we compared our norms with these studies to establish the validity for the current study. First, we compared the results with other Chinese studies (Zhang and Yang, 2003; Weekes et al., 2007; Liu et al., 2011). A comparison of the traditional line drawings (Liu et al., 2011) and the color images used in the present study (Table 6) revealed very small differences for all the common variables. Except for RT, percentage name agreement and concept agreement (p > 0.05), statistically significant differences were observed on other variables (p < 0.001). However, it is important to test the reliability of the differences across different studies, we conducted a correlation analysis on values of the overlapping variables to test whether the items elicits similar responses. We found that all these variables were significantly related to each other; the Pearson's correlations are presented in Table 7. The absolute values of the correlation coefficients were all high, suggesting that the items of the current study elicited similar response patterns to that of line drawings (Liu et al., 2011). The H-values in the current study were lower than those reported by Liu et al. (2011), which may suggest that color images have a more concentrated distribution of names.

Moreover, consistent with the results of Liu et al. (2011), most variables in the present study had significant correlations. We conducted factor analysis to provide more information about the internal structure of the correlated variables. Four factors were extracted from the ten variables: name agreement, lexical attributes of associated words, difficulty of image processing, and image agreement. The regression analysis further showed that familiarity, AOA, concept agreement, name agreement (%) and image agreement had a significant effect on naming latencies, which were strikingly similar to those reported by Liu et al. (2011). Weekes et al. (2007) also reported that name agreement (%), familiarity, AOA made independent contributions to naming RTs. The two most reliable predictors were name agreement and rated AOA, which were well documented for line drawings in different languages (Barry et al., 1997; Cuetos et al., 1999; Bonin et al., 2003; Nishimoto et al., 2005). In addition, we found that word frequency (log-transformed form, obtained from the SUBTLEX-CH dataset) had a significant impact on reaction time, which were in line with Zhang and Yang's study (2003) and other language studies (Barry et al., 1997; Cuetos et al., 1999). However, in Liu et al.'s study (Liu et al., 2011), the subjective word frequency had no significant effect on naming latencies. Instead of the subjective word frequency, we used the database of SUBTLEX-CH (Cai and Brysbaert, 2010) for word frequency (in log-transformed form), together with other variables in Liu et al.' study (Liu et al., 2011), to perform the simultaneous multiple regression analysis on naming RTs again. The total adjusted R² = 0.671, F_{(10, 423)} = 87.425, and p < 0.001. The results were consistent with the findings of Liu et al. (2011). Concept familiarity, objective AOA, rated AOA, concept agreement, and image agreement showed significant contributions to naming RTs, which were in the decreasing order of standardized beta coefficients. For the percentage of name agreement, there was a marginal significant (t = −1.867, p = 0.063). However, we found that word frequency had a significant contribution to naming latencies (t = −2.558, p < 0.05).

Previous studies have suggested that concept familiarity is equivalent to word frequency and that higher frequency corresponds to greater familiarity (Ellis and Morrison, 1998; Cuetos et al., 1999). These two variables may have the same effect on picture naming (Johnson et al., 1996; Liu et al., 2011). However, contrary to this view, some studies have suggested that they are very different. Almeida et al. (2007) found that familiarity affected object identification, whereas Jescheniak and Levelt (1994) found that word frequency affected phonological lexicon access. Moreover, Graves et al. (2007) revealed that activity in the left posterior superior temporal gyrus modulated by word frequency but not concept familiarity. Bates et al. (2003) investigated the object-naming norms for line drawings in seven languages and found that word frequency had large effects on naming latencies for all the languages. Moreover, in our study, factor analysis suggested that word frequency and familiarity were explained by different dimensions of the stimuli's characteristics. Therefore, as with other varibles (Name agreement, familiarity, concept agreement, AOA and object agreement), word frequency influenced the speed of picture naming.

For the colorized versions of line drawings, Bakhtiar et al. (2012) and Bonin et al. (2013) reported that name agreement, image agreement, and AOA norms made significant independent contributions to naming latencies. Furthermore, they found that objective word frequency was reliable in object-naming latencies. However, Bonin et al. (2013) found that subjective word frequency was no significant in object naming. For normalized photos of objects, Shao and Stiegert (2016) found that naming latencies were significantly predicted by word frequency, which obtained from the SUBTLEX-NL database in log-transformed form (Keuleers et al., 2010). Therefore, we suggest that the different operations of word frequency might affect its independent effects on picture naming. Objective word frequency might be more effective than subjective frequency.

Lastly, we compared the norms of the present study with other normative studies for color images, especially for those of Brodeur et al. (2010) and Moreno-Martínez and Montoro (2012). We compared the norms of the current study with those of Brodeur et al. (2010) for overlapping items (62 color images), which are shown in Table 8. Although common variables were normalized in different languages (Mandarin Chinese and English), only small differences between them were observed. However, a comparison of our data and those of Moreno-Martínez and Montoro (2012) for the 64 color images in Spanish (Table 9) reveals large differences in H-values and AOA. We conducted correlation analysis to explore the different values of variables across these studies and found that all these variables were significantly related to each other (see Table 7). Although the correlation coefficients for the H-value were both relatively low (r = 0.358, p < 0.01; r = 0.370, p < 0.01), the significant relations also suggest that the items of the current study elicited similar response patterns to those of Brodeur et al. (2010) and Moreno-Martínez and Montoro (2012). The higher H-value in the present study suggests that the given object has more possible names in Mandarin Chinese than in English and Spanish (Liu et al., 2011).

In summary, this is the first study to report norms for 435 color images in Mandarin Chinese and to normalize these images according to the following important variables: name agreement (H-value and percentage name agreement), concept agreement, word frequency, word length, AOA, familiarity, image agreement (object agreement and viewpoint agreement), and visual complexity. Sets of line drawings and sets of our color images could become complementary tools for experimental research. We expect that this work will enhance the usefulness of color images as materials for cognitive and psycholinguistic research. In combination with other existing sets of photos, we hope that this study will provide useful materials for further research across cultures.

QC conceived and supervised the experiments; QC and DZ designed the experiments; DZ implemented the experiments and collected data; DZ analyzed the results; and all the authors wrote and revised the manuscript.