Urban green space plays an increasingly important role in promoting human well-being [1
] and relieving stress and mental fatigue [7
]. Well-designed urban greening environment not only improves urban aesthetic quality [9
] but also promotes the physical and mental health of urban residents [10
]. Plants are the key environmental variable that form a major part of the urban green space [11
]. Previous studies on vegetation landscapes in urban green space have mainly addressed relatively large-scale vegetation landscape, such as urban forests [12
], parks [13
] and gardens [15
]. However, small-scale urban vegetation landscapes are particularly understudied in terms of the public’s aesthetic preference and emotional perception [16
In recent years, flower borders, a type of small-scale vegetation landscape, have been increasingly used in urban green space in China. Flower border is one of the application forms of flowers that combines flowers, herbaceous plants, shrubs and small trees [18
]. It is a natural ribbon flower arrangement with trees, hedges, low walls or architectures as the background. Flower borders were originally developed in English gardens and were widely used in parks, botanical gardens, private gardens and other places in America and many European countries. Compared with other forms like flower bed, flowering shrubs, flower belt, etc., flower borders have greater potential for increasing plant and landscape diversity [19
]. They can improve ecological benefit and maintain the stability of vegetation communities. A recent study on the public evaluation of urban flower borders revealed that flower borders with higher scores are characterized by high species diversity, rich colors and distinct landscapes [20
]. However, the visual landscape characteristics of flower borders have not been systematically and quantitatively studied, and it remains uncertain how these characteristics influence public aesthetic preference or emotional perception.
Previous studies on public subjective preference for vegetation landscapes have focused on characteristics such as foliage color, vegetation height and density, naturalness, and species richness [21
]. However, existing knowledge of preferred vegetation characteristics may not be applicable to predicting the preference and emotional response for flower borders because of their unique composition and characteristics. Color, as a basic aspect of human perception [24
], should be the most significant visual attribute that may influence emotional perception. Being the most changeable elements of nature and giving emotional perceptions are the characteristics of colors [25
]. Both foliage color and flower color play an important role in determining landscape preferences and perception [26
]. For example, due to the difference in emotional associations and visual quality, green-white foliage could develop negative feelings such as sadness and ugly, while green-yellow and bright green foliage could develop positive feelings like cheerful and active [27
]. In the case of flower borders, the collocation and combination of different colors may influence preferences and perceptions [28
]. In addition, Barnes states that cool colors (blues, greens, and purples), whites and pastel shades create a calming experience whereas hot and vibrant colors, such as bright reds, oranges and yellow, distract the mind through stimulation and are uplifting [29
]. Color features, including cool and warm colors [30
], and color brightness may play important roles in the visual attributes of flower borders. Likewise, color contrast has proven to be an important attribute of perceived beauty [13
]. Furthermore, the structural composition of flower borders, including their visual richness and the distribution of plant patches, may be important in determining preference and emotional response [7
]. This study seeks to identify the aesthetic preferences and emotional perceptions for different visual attributes of flower borders.
Measuring people’s perceptions and preferences has regularly used questionnaire methodologies that can be reliable to some extent [12
]. However, in terms of emotional perception, evaluation through self-reported scores has several apparent biases from the subjective emotion of respondents, real-time moods, problematic questions, and social role-restricted results [12
]. An objective method was used to obtain the emotional perception results in this study, which were then combined with the subjective preference results. Facial expressions represent an emotional response that provides a novel way to show people’s emotional perception when viewing vegetation landscapes. According to the circumplex model of affect [34
], all human reactions or affective states arise from two dimensions, valence and arousal. Valence is a measure of pleasantness (positive or negative) and arousal measures the level of activation of an emotion and indicates physical and mental alertness [35
]. This model has been applied to interpret human reactions to planting environments [32
]. Face reading is a novel technique that analyzes visual recordings of faces through a software algorithm that was generated by training the model using big data of intended emotion expressions [12
]. The current technique can achieve a facial analysis accuracy of up to 87% of the perceived emotion [36
], which provides the possibility to analyze the emotional perception of small-scale vegetation landscapes.
This study seeks to explore the public aesthetic preference for and emotional perception of flower borders in the context of the urban environment in China. We selected typical flower border images to explore the impacts of visual attributes on preference and perception. The questions to be answered include the following: (1) what is the relationship among the visual attributes of flower borders in urban green space; (2) what are the effects of the visual attributes of flower borders on people’s aesthetic preference and emotional perception; (3) what is the relationship between aesthetic preference and emotional perception; (4) what effect does the gender of participants have on aesthetic preference or emotional perception? The findings could contribute to a better design of flower borders for urban greening and provide an objective evaluation method for public emotional perception of vegetation landscapes.
2. Materials and Methods
Photographic images were used as surrogates for real flower border landscapes in this study as the effectiveness of this method has been widely confirmed [31
]. A total of 70 images of flower borders in urban green space were taken by one of the authors using a Sony Alpha 7R II digital camera between the hours of 10.00 a.m. and 2.30 p.m. on clear days. All images were transformed to a 3936 × 2214 px resolution using the Adobe Photoshop CS6 software. The photographs were not further digitally modified. To fulfill the objectives of this study, from all photographs, 18 final images that are the most representative scenes of flower borders were selected for evaluation according to the opinions of experts on landscape architecture (Figure 1
According to the psychophysical approach, landscape is a physical–visual experience [22
]. Quantifying the visual attributes of a landscape will improve the effectiveness of landscape evaluation, thus establishing the aesthetic preference and emotional perception for the landscape as a whole. Considering the characteristics of the samples, research of the literature [7
], and advice from experts on vegetation landscapes, nine visual attributes of vegetation landscapes were finally selected, as shown in Table 1
. We quantitatively evaluated the intensity of the visual attributes present in each image using categorical or continuous variables. Three categorical variables were assessed by the participants according to their perceptions using the questionnaire method (Table A1
). Six continuous variables related to plant color were measured using the ColorImpact 4 software based on the HSV color model [39
], which is adopted according to the human visual perception characteristics of color. We divided color hues into six categories according to the HSV color model, including red, yellow, green, cyan, blue and magenta.
A total of 113 participants (42 males and 71 females) aged between 18 and 52, were recruited from the Zhejiang University intramural forum for the internet surveys on aesthetic preference and 35 participants (13 males and 22 females) of them simultaneously participated in the face recognition experiment on emotional perception. The participants include college students, teachers and other school staff. The participants were carefully selected and did not have any mental illnesses, physical injuries, blindness, eye disease, color blindness or weakness.
The face recognition experiment took place in a quiet room at Zhejiang University. The room was set up specifically for the purposes of this study. Prior to the study, all participants were informed of the entire experimental procedure and the relevant equipment’s function within the experiment. All participants were volunteers and asked to read and sign a consent form if they wished to participate. At the beginning of the experiment, five ‘warm-up’ images were shown first as a preview sample of the study content. A series of pilot tests preceded the main study to define and fine-tune the experimental setup.
Participants were seated in a comfortable stable chair in front of a desk approximately 60 cm away from a 32″ full HD monitor with an adjustable height. Visual stimuli of 18 images were displayed at random. Each trial began with the display of a white image for 10 s, used as a baseline, followed by the display of the stimuli for 8 s; this process was repeated throughout the trial. Finally, the whole trial ended with an additional 10 s black image. A digital camera (Sony Alpha 7R II) was positioned to record the facial expressions of the participants above and behind the monitor.
After the face recognition experiment, each participant received an online questionnaire survey. The participants were asked to rate their aesthetic preferences for each image. They were instructed to rate the color brightness, color contrast and visual richness of the 18 images according to the measurement scale shown in Table 1
. There was no time limit to rate each image. When the ratings were finished, the demographic information of participants was requested. After the internet survey, each participant received 10 RMB as a reward.
2.4. Aesthetic Preference and Emotional Perception Measurement
In this study, the participants’ aesthetic preferences for each image of flower borders were collected through a questionnaire survey. This method has been widely used by previous researchers, and its reliability has been generally accepted [40
]. We measured the aesthetic preference for images of flower border as the degree to which participants agreed with the statement ‘this vegetation landscape is very beautiful’. Participants were asked to rate the 18 images on a 5-point rating scale ranging from 5 = strongly agree to 1 = strongly disagree (Table A1
). The ratings were used as the dependent variable representing aesthetic preference in the post hoc analysis.
Facial expression measurements were used to generate the valence and arousal of the participants when viewing the 18 images of flower borders. The valence measures the level of pleasure while the arousal indicates the level of affective activation [42
]. Based on the large-scale Aff-Wild2 dataset and the Affective Behavior Analysis in-the-wild, we used the stat and temporal module to fine-tune face features again to establish the emotion expression analysis model [43
]. More details about the mechanism and training methods of the model can be found in Do Nhu and Kim [43
]. This measure analyzes the videos of participants’ facial reactions during the experiment frame-by-frame and predicts the emotion with small displacements in the faces. The model eventually outputs the valence and arousal of each participant for each frame. We calculated the average valence and arousal of each flower border image and white image to obtain the emotional fluctuation of the participants when viewing the flower border landscapes. The values of the changes in valence and arousal were used as two dependent variables related to emotional perception in the post hoc analysis.
2.5. Statistical Analysis
All statistical analyses were performed using SPSS 20 software. The mean value for each visual attribute in each image was calculated and recorded. To abstract the main characteristics of the urban flower border landscapes, principal component analysis (PCA) was performed using the normalized varimax rotation procedure. This method allows redundancies among visual attributes to be detected and reveals the potential dimensions of the data. One-way analysis of variance (ANOVA) and correlation analysis were employed to examine the differences among emotional perception, including valence and arousal, and aesthetic preference with various visual attributes of the flower borders. We performed Levene’s test of equality of variances to quantify the homogeneity of the variance. A general linear model was used to explore the significant predictors of aesthetic preference and emotional perception. The final minimum adequate models were obtained via the backward elimination of nonsignificant (p > 0.05) variables. Correlation analysis was then used to identify associations between aesthetic preference and valence, arousal and aesthetic preference, and valence and arousal. In addition, an independent-sample t-test was used to examine the differences in aesthetic preference, valence and arousal in different gender groups of respondents.