Urban Architectural Color Evaluation: A Cognitive Framework Combining Machine Learning and Human Perception

Abstract

Architectural color significantly impacts the quality of built environments and is closely related to the physical and mental health of residents. Previous studies have conducted numerous valuable explorations in this field; however, the challenge of quantitatively measuring the characteristics of architectural colors in depth and examining the complex relationship between these colors and human perception remains an unresolved issue. To this end, this study builds upon recent advancements in data technology and emotion analysis to develop a comprehensive cognitive framework for urban architectural color evaluation. It combines machine learning techniques and perception scales, utilizing both objective and subjective data. By acquiring and recognizing numerous street-view images of the Changsha Central District, we quantitatively examined the hue, saturation, value, color complexity index and color harmony index of urban architectural colors and investigated the complex relationships between human perception and architectural colors through large-scale participant ratings and correlation analyses. The results show that the architectural colors of the study area are warm, with low saturation and moderate value. Most areas exhibit a high color complexity index, whereas the overall color harmony score is low. Human-perception evaluations indicate that people are generally satisfied with the urban architectural colors of the Changsha Central District. For further optimization, the saturation and color harmony scores need to be enhanced. This study provides a comprehensive evaluation of urban architectural color quality, visualizing the complex relationship between urban architectural color and human perception. It offers new perspectives for improving the built environments and supporting sustainable development, with practical application value.

1. Introduction

Owing to the constant acceleration of urbanization, China’s urbanization rate is expected to reach 75–80% by 2035 [1]. In recent years, public demand for urban spaces has shifted from merely pursuing quantitative growth to focusing on high standards and quality. This has considerably increased the importance of built-environment quality, as well as the livability and spatial quality of cities in sustainable urban development strategies [2]. Moreover, architectural color is known to significantly impact the emotions, cognition, and behavior of residents [3]. Architectural color is an objective phenomenon that represents the reflection of light from the surface of a building onto the human eye and is considered the most intuitive factor by which individuals evaluate the quality of a built environment, which significantly impacts their environmental experience [4]. The significance of “Double Urban Repairs” for urban transformation and development was emphasized in both the Opinions of the Central Committee of the Communist Party of China and the State Council on Accelerating the Construction of Ecological Civilization in 2015, as well as the Several Opinions of the Central Committee of the Communist Party of China and the State Council on Further Strengthening Urban Planning, Construction, and Management in 2016. The concept of “City Betterment” focuses on urban architectural color and has led to the development of various guiding principles and practical investigations [5]. Despite the growing recognition of the significance of urban architectural color in contemporary urban planning, as demonstrated by the implementation of the “Changsha Urban Color Planning” and the “Regulations on the Administration of Urban Architectural Colors in Changsha” in 2009, which represent significant progress in guiding and regulating urban architectural colors, their practical implementations frequently lack specific guidance or overly prioritize rigid uniformity, neglecting place experiences that are most relevant to urban residents [6]. To promote sustainable urban development, it is crucial to offer guidance and develop practices regarding urban architecture colors.

Architectural color has attracted considerable attention in theoretical discussions and practical applications as a significant characteristic of architecture [7]. As research has progressed, the importance of architectural colors in large-scale environments has become increasingly prominent. This factor plays a crucial role in shaping urban characteristics, preserving historical context, and stimulating creative ideas for urban planning, heritage conservation, and spatial revitalization. Additionally, research on architectural color is gradually promoting the transformation of urban design concepts from focusing solely on visual beauty to comprehensive considerations of cultural heritage, environmental psychology, and social identity [8,9,10,11].

Recent technical advancements have led to the creation of new data environments, broadening the measurement methods and application scope of urban architectural colors. Machine-learning-based street-image recognition technologies have enabled the automatic collection and analysis of large-scale color data. This advancement has significantly enhanced the processing capacity and analytical precision of urban color data, thereby offering a scientific framework for color planning and assessments in urban areas [4,12,13,14]. Additionally, the trend of interdisciplinary research has become increasingly prominent in recent years, with many scholars dedicated to bridging the subjective and objective realms by quantifying the influences of the constituent elements of large-scale urban areas on human perception through technological methods. These approaches aim to analyze the roles of psychological, social, and cultural factors in the material world more comprehensively and provide more precise evaluations [15,16,17,18,19,20]. Additionally, some studies have identified more profound influential factors, such as personal preferences, cultural backgrounds, and economic differences, that are crucial for comprehensively interpreting human perceptions of built environments [21,22,23].

Thus, current studies are aimed at developing more sophisticated and systematic analytical frameworks for comprehensive evaluations of architectural color through data analysis and perceptual evaluations. Machine-learning-based street-view image recognition is widely used for extracting architectural colors, which are subjectively evaluated using psychological scales and physiological indicators. An increasing number of studies have suggested that the concept of architectural color involves multiple fields such as physics, psychology, and art, and analyzing it both subjectively and objectively can not only enhance our understanding of the quality of built environments [2,24] but also help determine the psychological feelings and behaviors of those affected by it [25,26,27]. Such analyses can also provide theoretical and practical guidance for urban planning and architectural designs [2,25] and their importance has been widely recognized [28]. Owing to these continuous technological advancements, architectural color assessments are expected to be refined further in the near future. This will provide more scientific and efficient support for creating more livable and sustainable built environments, thereby promoting the psychological health and social well-being of residents.

However, based on the analysis of existing research, we found that the current quantitative analysis methods for recognizing and extracting architectural colors solely from street view images have limitations, as it is difficult to uncover deep level color features and combinations. A lack of in-depth quantitative methods will undermine the credibility of the analysis of the complex relationship between architectural color and crowd perception. Additionally, when evaluating human perception, the subjectivity inherent in the evaluation process can result in significant data bias due to the single dimension of the evaluation criteria. Therefore, this study optimized and improved the current methods for measuring architectural colors. Specifically, machine learning techniques were used to recognize architectural colors in a large dataset of street view images, and calculation formulas were developed based on the relative relationships between color pixels to measure objective color data more accurately. Furthermore, the advantages of the Perceived Restorativeness Scale (PRS) and Semantic Differential (SD) analysis were integrated to design a human perception scale for evaluating architectural colors. This approach combines refined indicators with in-depth interpretation, providing a more accurate explanation of human perception. Through correlation analysis of objective and subjective data, this study comprehensively explores the color quality of urban buildings in the research area and the psychological perceptions individuals have regarding architectural colors.

2. Materials and Methods

The cognitive framework proposed in this study for urban architectural color evaluation is based on a quantitative analysis of the correlation between objective color data obtained through machine learning and subjective color evaluation using perception scales.

Machine learning can extract valuable feature information from large, high-dimensional datasets with remarkable precision and efficiency. Specifically, it can effectively analyze complex spatial environment information through learning from big data, enabling precise capture and deep understanding of the structure, atmosphere, and quality of the built environments. This capability allows machine learning to master core features that significantly surpass human abilities. In existing research, machine learning techniques have been validated for their accuracy and efficiency, becoming the primary method for assessing the degree of environmental greening, interface attractiveness, and architectural colors [4,12,13,14,15,16,17,18,19,21,22,23,24]. Additionally, perception scales can represent subjective feelings about the built environments as data and visualize them in an intuitive manner, thereby simplifying the analysis process. Previous studies have utilized PRS to measure the subjective evaluation of the built environment by crowds. The diverse indicators provided by PRS can effectively differentiate various subtle emotional perceptions, resulting in precise subjective evaluation results. Furthermore, SD analysis, with its excellent descriptive ability, facilitates individuals in articulating their subjective feelings and has also been applied in research [23,25,29,30,31,32,33,34,35,36]. Based on the research presented above, we proposed the conceptual framework and research objectives of this study, as illustrated in Figure 1.

2.1. Study Area

This study employed Changsha Central District as the study area. First, as the capital city of Hunan, which has a high level of urbanization, Changsha provides convenient access to urban architectural color samples. Second, as it is situated in a hilly region featuring hot summers and cold winters, it boasts a rich historical heritage and rapid development, resulting in diverse architectural styles influenced by its natural and cultural environments. Additionally, Changsha was an early adopter of color planning in China and can therefore serve as a good representative case. Finally, Changsha Central District comprises diverse urban areas, including a central city, university towns, and parks, rendering it ideal for analyses.

2.2. Acquisition of Street View Images

We obtained the vector road network data of Changsha from OpenStreetMap (OSM). To simplify the road network, broken, incomplete, and misidentified roads in the data were eliminated using the ArcGIS 10.8.1 software. Additionally, considering the effective sight distance of pedestrians on flat city streets, a minimum distance of 100 m was set between sampling points. Subsequently, 13,185 sampling points, along with their respective latitude and longitude coordinates, were randomly generated on the road network.

The Baidu Map API was used to retrieve 2048 × 664-pixel panoramic street-view images in batches using specific point coordinates. Images of sampling points not covered by Baidu Street View and those with disproportionately low architectural elements were discarded to ensure accuracy. Finally, a dataset comprising 10,396 panoramic street-view images was generated (

Table 1. Examples of panoramic street view images.

Point	Point_X	Point_Y	Image
1	112.9476644	28.19954668
2	112.9487911	28.19950462
3	112.9497763	28.19949366
4	112.9506871	28.19946451
5	112.9546989	28.14211017
……
10,396	112.9227749	28.16553636

). Figure 2 shows a panoramic street-view image obtained through this process.

2.3. Acquisition of Color Features

In this study, the dominant color was used to evaluate the urban architectural color. Dominant color refers to a set of primary colors in a particular context that effectively and intuitively mirrors the characteristics and general inclination of a color palette. The dominant color in urban architecture plays a critical role in shaping a city’s visual identity, thereby directly influencing the overall quality of the built environment. It is one of the principal media through which individuals perceive urban architecture, rendering it a significant and viable research domain [37].

2.3.1. Hue, Saturation, and Value

The color data used in this study were based on the hue, saturation, and value (HSV) color space, which closely correspond to the human visual perception of colors, thereby aiding their processing and analysis. Hue represents the position of a color on the spectrum, measured in degrees within the range of 0–360°, wherein those from 0–90° and 270–360° are categorized as warm tones, whereas those from 90–270° are categorized as cool tones. Saturation is a measure of color purity and ranges from 0–100%, where higher values represent more intense and pure colors, whereas lower values tend toward gray. Finally, value signifies the luminance of a color and ranges from 0–1, with higher values indicating brighter colors, whereas lower values indicate darker colors.

We employed the segmented neural network (SegNet) model, a convolutional neural network (CNN) architecture, in PyCharm to semantically segment the 10,396 panoramic street-view images and extract architectural elements. In the pre-preparation we compared CNN with PSPNet, Unet, and other methods and verified its more accurate results. Subsequently, the CameraRaw tool in Photoshop (Adobe) was employed to perform batch white-balance adjustments on the images to eliminate color distortions caused by environmental factors such as weather and time of day.

We adopted a color extraction method based on the K-means clustering algorithm and used the architectural images as inputs. Clustering was performed based on the position and distance of the pixels in the color space to divide it into three clusters, each representing a dominant color. The HSV data and percentages of the three dominant colors in the images were obtained using the algorithm. The process of acquiring and analyzing the image elements is illustrated in Figure 3.

2.3.2. Color Complexity Index

Previous studies have highlighted that the complexity of an image color cannot be adequately captured solely based on the number of colors in it, which may result in errors by overlooking all of the colors therein. Therefore, Qi et al. [26] proposed the concept of color complexity, which calculates the proportion of colors in an image as follows:

$$C_C=|\left({\displaystyle \sum }_{i=0}^n{S}^2\right)-1|$$

(1)

where $C_C$ represents the color complexity score, and $n$ and $S$ denote the number of dominant colors and percentage of each dominant color in the image, respectively. In order to strike a balance between the accuracy of results and ease of calculation, the number of dominant colors was set to 3. The color complexity score ranges from 0–1, with lower values indicating simpler overall color in the image.

2.3.3. Color Harmony Index

Color complexity analysis typically focuses on the quantity and proportion of the colors in an image. However, the harmony between these colors also significantly influences human perception [38]. This study presents a formula for calculating color harmony score using color data extracted from images. The equation considers the properties of the HSV color space, assesses the consistency between different pixels by calculating their relative distances within the color space, and subsequently determines the overall color harmony. After comparing this approach with existing studies [26], it maintains suitable computational effort and high research efficiency while basically preserving the color characteristics, as shown in Equation (5).

$$\Delta H={\displaystyle \frac{{{\displaystyle \sum }}_{i=1}^n{{\displaystyle \sum }}_{j=i+1}^n|H_i-H_j|}{\left(\begin{array}{c}n\\ 2\end{array}\right)}}$$

(2)

$$\Delta S={\displaystyle \frac{{{\displaystyle \sum }}_{i=1}^n{{\displaystyle \sum }}_{j=i+1}^n|S_i-S_j|}{\left(\begin{array}{c}n\\ 2\end{array}\right)}}$$

(3)

$$\Delta V={\displaystyle \frac{{{\displaystyle \sum }}_{i=1}^n{{\displaystyle \sum }}_{j=i+1}^n|V_i-V_j|}{\left(\begin{array}{c}n\\ 2\end{array}\right)}}$$

(4)

$$C_h={\displaystyle \frac{n}{\Delta H+\Delta S+\Delta V+1}}$$

(5)

where $C_h$ represents the color-harmony score; $n$ represents the number of dominant colors in the image; and $H_i$, $S_i$, $V_i$, $H_j$, $S_j$, and $V_j$ represent the HSV data for the three dominant colors in the same image. Additionally, $\Delta H$, $\Delta S$, and $\Delta V$ correspond to the differences in hue, saturation, and value, respectively. A higher color harmony score indicates a more harmonious color composition.

2.4. Development of Perceptual Scales

The PRS, originally formulated by Hartig et al. [29] by drawing upon Kaplan’s attention restoration theory, assesses the extent to which environments support the restoration and replenishment of the mental and physical resources of individuals [30,31,32]. It encompasses multiple factors such as being-away, fascination, extent, and compatibility. It gathers the subjective experiences and perceptions of participants through a series of questions to quantify the restorative potential of the environment. Over time, scholars have refined and improved the PRS through empirical applications.

SD analysis, as defined by Osgood, is a psychometric technique that utilizes natural language processing to quantify psychological evaluations [33]. It was initially widely employed in psychology and has subsequently been expanded to various disciplines, such as architecture, urban planning, and landscape design. It involves comprehensive spatial perception assessments that quantitatively capture the experiential aspects of an environment [25,34,35,36].

The perception scale is developed based on the PRS scale and SD analysis, following this process: Firstly, we randomly selected 30 participants from both online and offline sources, conducting nearly 3 h of interviews in total. These interviews covered topics such as emotions, activities, and events related to urban architectural colors. Participants were not chosen from specific demographic groups to ensure a more accurate representation of public perceptions. Subsequently, we conducted a word frequency analysis of the interview content to comprehensively classify and summarize the impact of architectural color on human perception. Specifically, the public’s emotional perception of architectural colors includes 7 indicators: Sense of Character, Sense of Vitality, Sense of Security, Sense of Cleanliness, Sense of Belonging, Sense of Affinity, Influence. Intuitive evaluation includes 3 indicators: hue, complexity, harmony. Finally, we expanded the 10 indicators into 10 descriptive questions that provide context, pre-set answer options and corresponding scores. Each criterion was evaluated on a five-point scale with a possible maximum score of 50 points, allowing participants to make the most accurate choices when presented with random street view images (Figure 4,

Table 2. Items in the human perception scale.

Indicators	Subjects	Keywords
Sense of Character	The colors in the city appear:	Fuzzy–Unique
Sense of Vitality	When I approach the architectural colors, I feel:	Bored–Lively
Sense of Security	When I am surrounded by the architectural colors, I feel:	Anxious–Relaxed
Sense of Cleanliness	The architectural colors give me the impression that the city is:	Polluted–Healthy
Sense of Belonging	The architectural colors can evoke my memories and imagination.	Inconsistent–Consistent
Sense of Affinity	I want to stay in the architectural environment for a longer duration.	Inconsistent–Consistent
Influence	I think different architectural colors will change my mood.	Inconsistent–Consistent
Hue	I like the tones presented in the images.	Inconsistent–Consistent
Complexity	I think the colors are very rich.	Inconsistent–Consistent
Harmony	I think the colors are very harmonious.	Inconsistent–Consistent

).

In this study, we conducted a complete experimental process. Firstly, sixty street view images showing a significant proportion of architecture with clearly distinguishable colors were randomly selected. To minimize experimental errors resulting from randomness, 3 identical perceptual scales were provided for each image, resulting in 180 scales. Each scale will be scored and evaluated individually by different participants. Subsequently, participants were selected based on the following criteria: (1) sufficient visual and color recognition abilities, (2) residence in urban areas with a high density of buildings, and (3) a habit of observing and interacting with the built environment. The final study sample comprised 180 urban residents from various regions of the country, including 102 males (56.67%) and 78 females (43.33%), with a mean age of 24.87 years. Of these, 30 had educational qualifications in architecture. Finally, after comprehending the experimental objectives and process, participants completed the perception scales and rated the street view images assigned to them. Each street view image received a collective evaluation from 3 participants.

3. Results and Discussion

The results are based on the above research processes to analyze both objective and subjective data, as well as their correlations. They are divided into four sections: visualization of dominant colors, quantitative characteristics of colors (including numerical analysis of HSV, color complexity index, color harmony index), evaluation of crowd perception, and research progress. This structure aims to comprehensively and in depth evaluate urban architectural color, visualize the complex relationship between urban architectural color and human perception, and provide new perspectives for improving the built environments and supporting sustainable development.

3.1. Visualization

The dominant colors of the urban architecture in the Changsha Central District, as identified through K-means clustering, are outlined in

Table 3. Examples of architectural color visualization.

Point	Image	Dominant Colors	Classification	Features
1			High-rise office buildings; Structures	Modern, Stable Practical, Durable.
2			High-rise office buildings; Multistory exhibition buildings	Modern, Stable; Technological, Unique.
3			High-rise office buildings; Multistory public buildings	Modern, Stable; Flexible, Suitable.
4			High-rise office buildings; Multistory exhibition buildings; Multistory residential buildings	Modern, Stable; Technological, Unique; Warm, Comfortable.
5			High-rise residential buildings; Multistory office buildings	Modern, Stable; Professional, Economic.
……
10,396			Low-rise commercial buildings; Low-rise residential buildings	Distinctive, Lively; Friendly, Harmonious.

. The colors in the built environment within a 360° range of the sampling point were succinctly and efficiently represented to capture intuitive perceptions. The architectural colors in the study area exhibited rich and diverse characteristics. Specifically, residential buildings, school buildings, and traditional neighborhoods were predominantly characterized by warm tones, fostering a cozy and comfortable atmosphere, and conventional commercial areas frequently employed bright and vibrant colors to cultivate lively and attractive commercial settings. In contrast, contemporary commercial areas and high-rise buildings exhibited cool tones, highlighting their modern and prosperous attributes. The Science and Education District included uniform and light colors to establish an appropriate environment for academia and research. Ancient architectural designs emphasized the importance of color harmony and unity, effectively highlighting their unique historical and cultural features.

The analysis revealed that the diversity of dominant colors at the sampling points directly reflects the variety of urban architectural colors in the Changsha Central District, showcasing the vibrant appearance of its architecture. The spatial distribution of dominant colors demonstrates a specific pattern intricately linked to the architectural style, functional placement, and historical and cultural contexts of diverse regions. The amalgamation of colors significantly influences the aesthetics of architecture and can generate diverse atmospheres within the environment. This information is important for enhancing the comprehension and strategization of urban color environments.

3.2. Architectural Color Characteristics

3.2.1. Numerical Analysis

The hue data for the 10,396 sampling points used in this study are presented in Figure 5. Evidently, 57.83% of the points exhibited dominant colors in warm tones, whereas 42.17% exhibited dominant colors in cool tones. The saturation data presented in Figure 6 indicate that all data points were ≤0.6, and 64.03% were concentrated within the range of 0–0.1. The number of points decreased as the saturation increased. The color value data presented in Figure 7 indicate that all points were within the range of 0.3–0.9, with 90.89% concentrated between 0.4 and 0.6.

The analysis revealed that the Changsha Central District predominantly comprises architectural colors with warm tones. Additionally, the overall saturation was relatively low, with occasional accents of high saturation. The general values fell within a moderate range, contributing to the simple, elegant, and bright appearance of the urban architecture.

3.2.2. Color Complexity Index

All of the numerical features were calculated for each data point, and the results are listed in

Table 4. Examples of color complexity calculations.

OID	Dominant Color 1-PCT	Dominant Color 2-PCT	Dominant Color 3-PCT	Cc
1_112.9476644_28.19954668	0.428315601	0.391586783	0.180097616	0.630770386
2_112.9487911_28.19950462	0.411126753	0.300954169	0.287919078	0.657503986
3_112.9497763_28.19949366	0.447991154	0.314780686	0.237228161	0.643939845
4_112.9506871_28.19946451	0.579630751	0.286228814	0.134140436	0.564107602
5_112.9546989_28.14211017	0.581900669	0.269420567	0.148678765	0.566698794
……
10396_112.9227749_28.16553636	0.483649331	0.258520059	0.257830609	0.632774081

. The data indicate that most points (79.93%) exhibited a color complexity between 0.6 and 0.7, suggesting that urban architectural colors were generally diverse across the study area. However, certain areas exhibited monochromatic color palettes.

The visualization of the results in ArcGIS, as shown in Figure 8, demonstrates that areas with more intricate color patterns tend to aggregate, particularly in scenic areas, commercial districts, residential zones, and along transportation routes. Conversely, areas characterized by lower color complexities exhibit a more dispersed distribution and are mainly those around schools, traditional neighborhoods, and newly developed regions. Additionally, the urban architectural colors in the Changsha Central District tend to be more diverse in high-vitality areas. By contrast, large-scale units, newly constructed areas, and historical districts typically exhibit more uniform and less diverse color compositions.

3.2.3. Color Harmony Index

The numerical characteristics of all sampling points were calculated, and the results are presented in

Table 5. Examples of color harmony results.

OID	H	S	V	Ch
1_112.9476644_28.19954668	160  200  206.6666667	0.075  0.012820513  0.072580645	0.156862745  0.917647059  0.48627451	0.091856174
2_112.9487911_28.19950462	210  180  200	0.034334764  0.1  0.083916084	0.91372549  0.235294118  0.560784314	0.13956043
3_112.9497763_28.19949366	200  105  150	0.026086957  0.102564103  0.030075188	0.901960784  0.152941176  0.521568627	0.046236599
4_112.9506871_28.19946451	210  214.2857143  213.3333333	0.05  0.115384615  0.113924051	0.941176471  0.71372549  0.309803922	0.694179685
5_112.9546989_28.14211017	216  210  60	0.021834061  0.065789474  0.025	0.898039216  0.596078431  0.31372549	0.02845791
……
10396_112.9227749_28.16553636	9.230769231  46.66666667  18.11320755	0.541666667  0.043269231  0.346405229	0.188235294  0.815686275  0.6	0.112326607

. The color harmony scores of most points (48.48%) were concentrated in the range of 0–0.1. As the score increased, the number of points gradually decreased. A comparison between the range of coordination degrees exceeding one and the range of the lowest coordination degree indicated a notable decrease in the number of points in the former, constituting approximately 4.05% of the latter. The analysis revealed significant variations in the architectural colors employed in the study area, resulting in an overall disordered visual impression.

The results were visualized using ArcGIS, as shown in Figure 9, wherein points exhibiting a lower color harmony score are densely clustered and dispersed throughout multiple areas within the Changsha Central District. Additionally, those exhibiting a higher color harmony score are scattered across various locations, primarily plazas, schools, high-rise areas, transportation routes, and riverside landscapes. Therefore, the urban architectural color scheme of the Changsha Central District lacks harmony, leading to a visually chaotic appearance, and only a few regions subjected to formal planning exhibit a satisfactory degree of color coordination.

3.3. Human-Perception Evaluation

In the experimental design of this study, subjective data were used to compensate potential discrepancies between machine learning outcomes and actual human perception. However, during the course of the study, we further discovered that the data can be used as supporting material for the accuracy of machine learning in capturing color features. In total, 180 data points were collected in this study. All participants rated the randomly presented images after gaining an understanding of the research, and the results are illustrated in Figure 10, which was visualized using the Origin 2021 software. The color data corresponding to each image were retraced after the ratings were completed. Considering the complex relationship between variables, the results were subjected to Spearman’s correlation analysis using the Statistical Package for Social Sciences (SPSS), and

Table 6. Correlation analysis between objective data and subjective perceptions.

Spearman’s Correlation Analysis
	Sense of Character	Sense of Vitality	Sense of Security	Sense of Cleanliness	Sense of Belonging	Sense of Affinity	Influence	Hue	Complexity	Harmony
H	0.023	−0.190	−0.475 **	−0.124	−0.156	−0.585 **	−0.722 **	−0.624 **	−0.216	−0.537 **
S	0.130	0.314 *	0.315 *	0.180	−0.039	0.306 *	0.152	0.138	0.085	0.367 **
V	0.193	−0.018	0.183	0.287 *	0.055	0.204	−0.054	−0.077	−0.156	0.248
Cc	0.007	0.588 **	0.216	0.240	−0.186	0.034	0.425 **	0.231	0.556 **	0.055
Ch	0.060	0.160	0.693 **	0.331 **	0.175	0.682 **	0.339 **	0.338 *	−0.032	0.788 **

* p < 0.05; ** p < 0.01.

presents the correlations between the perceptual indicators and architectural color characteristics.

The results indicate an average total score of 30.86 across all scales, with 66.67% of the images exhibiting a cumulative score of ≥30. The average score for images predominantly featuring warm colors was 32.30, whereas that for those predominantly showing cool colors was 28.83. Among the scales with total scores of ≥30, the average saturation data (0.13) were higher than the overall mean saturation data (0.09) for architecture within the study area. Additionally, the mean value (0.50) was the same as the overall mean value (0.50). The difference between the average color complexity score (0.65) and the general mean (0.62) was minimal, whereas the average color harmony score was 0.33, surpassing the overall mean of 0.20. Additionally, seven of the ten evaluative indicators exhibited a stronger correlation with at least one of the color feature datasets, one exhibited an insignificant correlation, and two exhibited low correlations.

The analysis revealed that people in the Changsha Central District are satisfied with the overall urban architectural color and prefer warm tones, moderate color complexity, and high color harmony. Improving the saturation and color harmony is essential for color quality, as architectural color and human perception are correlated. Subsequently, we conducted a semantic analysis-based pointwise retrospection and curve fitting of the scale statistics. Based on the compilation of the perceptual evaluation results with scores above 30, which indicate a high level of satisfaction with architectural colors, the preferred color characteristics identified by the people are as follows: H ∈ (0, 50), S∈(0.17, 0.27), V∈(0.47, 0.52), Cc∈(0.58, 0.60), Ch∈(0.60, 1.20). This finding will be used to evaluate whether the urban architectural color falls within the range that satisfies the people.

3.4. Progress in Research

This study extends previous research for analyzing architectural colors. We employed novel data-processing and perception evaluation techniques to precisely evaluate architectural colors and understand human perceptions by collecting and identifying a large number of street-view images and conducting a questionnaire survey. The advancements made in this study can be summarized as follows: (1) Previous studies have primarily concentrated on refining color recognition algorithms and analyzing images based on dominant colors [4,12,22,24,37]. In contrast, this study further extracted numerical features to calculate color complexity and harmony indicators derived from the above studies, thereby enhancing color recognition accuracy and efficiency. (2) Previous studies have emphasized the importance of subjective measures and have developed several perceptual scales [23,25,26,28,35]. In this study, we built upon this foundation by further integrating the PRS and SD analysis to enhance the assessment dimensions, experimental credibility and efficiency, and crowd evaluation accuracy. (3) We also conducted a correlation analysis of subjective and objective data to better understand the intricate correlations between architectural color quality and psychological perceptions of humans as shown in Figure 11. Additionally, the reliability of the experimental findings was validated.

This study confirms and supplements previous research results on architectural color: (1) Architectural color is one of the most intuitive visual elements of the built environments, significantly influencing both the quality of these spaces and the emotional experiences of individuals [9,12,25,26,28]. (2) Due to the objective nature of architectural color and its impact on emotions, it is essential to evaluate it using a combination of subjective and objective approaches [9,12,26]. (3) Based on the research findings presented above, a more detailed analysis of the quantitative characteristics of architectural color, along with the development of a perception scale featuring a multidimensional evaluation system, will contribute to the establishment of a cognitive framework for urban architectural color evaluation. This approach will ultimately enhance the accuracy and efficiency of future research.

Additionally, our findings can be summarized as follows: (1) In the Changsha Central District, urban architecture predominantly features warm color tones and low overall saturation. High-saturation colors are used locally for embellishment, and their overall brightness falls within a moderate range, imparting an elegant and bright urban architectural appearance. In terms of personalized color selection and overall color scheme, most areas are complex and lack harmony. Based on the specific situation of the study area, we have proposed responses after analyzing each indicator. (2) Architectural color is directly influenced by factors such as color planning, the involvement of relative institutions, spontaneous construction impacts, and potential influences from historical, cultural, and economic perspectives. (3) The HSV data of architectural color, along with the color complexity and color harmony scores calculated based on it, affect human perception. Specifically, 5 characteristics of urban architectural color (H, S, V, Cc, Ch) exhibit notable and complex correlations with 10 typical preference points of feedback from human perception (as shown in the perception scale). This understanding is crucial for urban and architectural designs.

4. Conclusions

The framework proposed in this study provides a method for evaluating the quality of the built environment and guiding practical design applications. The specific application scenarios are as follows. (1) The evaluation of existing architectural colors can be conducted by combining machine learning to measure the quantitative data of architectural colors (H, S, V, Cc, Ch) and crowd evaluations obtained from perception scales, which is also the approach adopted in this study. Alternatively, based on machine learning to obtain quantitative data, the partitioning of reasonable ranges of H, S, V, Cc, and Ch in this study can also be directly used to evaluate the urban architectural color. (2) In the context of urban renewal and renovation, the perception scale developed in this study can be directly used to obtain subjective evaluations. Based on the correspondence between the scale content and the characteristics of architectural colors, the population’s preferences can be determined, enabling targeted renewal and renovation to be carried out. The accuracy of the scale has been validated through research. (3) In future urban architectural design, the appropriate ranges of H, S, V, Cc, and Ch in a project can be determined by evaluating quantitative data on the colors of the surrounding environment. This evaluation will guide the selection of materials and colors during the design process. Additionally, continuous feedback on crowd evaluations through perception scales in the actual construction process will help improve design quality.

As a data-driven approach, our study demonstrates its generalizability and utility through the following case (Figure 12). We selected a street view image with low participant scores, a traditional neighborhood in the Changsha Central District. The goal is to enhance the quality of the built environment by altering the architectural color in the central section of the image. Firstly, the following two types of color data for the image were backtracked: the HSV data, color complexity index, and color harmony index of the entire image; and the HSV data of the three dominant colors in the image along with their respective percentages. Subsequently, referring to Equations (1) and (5), and the people preference intervals for Cc and Ch data summarized in Section 3.3, the appropriate direction for adjusting the HSV data of the entire image can be obtained. Finally, we modified the current color in Photoshop and conducted a second round of scoring on the image, with the same three participants scoring significantly higher. In practice, after calculating the numerical optimization directions for the five indicators, relevant building materials or coatings can be selected for design based on their characteristics. This suggests that the cognitive framework for urban architectural color evaluation provided according to this study can be effectively utilized in the fields of architectural and urban design. Additionally, although this case primarily focuses on the optimization and renovation of existing architecture, its principles are also applicable to the comprehensive planning of new constructions.

However, this study had several limitations, which can be addressed in future studies. (1) Street-view images from Baidu Maps have several issues, including blurry images, static positions, varying shooting angles and pedestrian viewpoints, and outdated images. Therefore, in future research, we will strive to establish a more real-time and precise urban architecture database by extensively utilizing images from social media and those captured by our team. This approach will expand our data collection channels and avoid the impact of image quality on human perception. (2) When calculating color features, the variations in architectural facade decorations and reflection of the surrounding environment on glass curtain walls can influence the recognition of inherent architectural color. Additionally, the color-calculation formula has some limitations. Future research should explore more advanced image-recognition technologies, such as optimizing the architecture of CNN and K-means clustering, as well as the development of new algorithms capable of intelligently identifying and excluding temporary elements on building facades. This approach aims to enhance the accuracy of color recognition. (3) During subjective evaluations, factors such as a limited number of questions, finite number of scales, and the inability of planar images to recreate realistic scene sensations affected the stability and reliability of the results to a certain extent. In future research, we plan to expand the scope of participants to minimize subjective data errors. Simultaneously, we will incorporate Virtual Reality (VR) technology to create a roaming virtual space that offers a more immersive experience. This approach will enable participants to identify environmental elements more accurately and enhance the credibility of models related to environmental emotions.

Urban architectural color has garnered increasing attention in the context of contemporary urbanization as an intuitive representation of urban style and a crucial visual element influencing human perception of the built environment. It not only represents the overall image of urban architecture but also influences the living experiences and emotional well-being of residents. A pleasant urban color environment has the potential to elevate the well-being and contentment of inhabitants as well as foster economic growth and the long-term viability of the city. Therefore, future urban and architectural design endeavors must focus on the quality of urban architectural color. Furthermore, new technologies should be integrated to foster the development of more aesthetically pleasing, cohesive, and habitable built environments.