A Method for Measuring the Visual Coherence of Buildings in Residential Historic Areas: A Case Study of the Xiaoxihu Historic Area in Nanjing, China

Abstract

Residential historic areas are currently the main focus of urban renewal efforts in China, primarily consisting of traditional residential buildings with similar characteristics. Hence, the visual coherence of buildings (VCoB) plays a crucial role in such areas, not only pertaining to the visual quality of the urban landscape but also regarding the preservation of historic features. The accurate measurement of the VCoB is a prerequisite for undertaking optimization efforts. However, discussions of the VCoB in the built environment are limited and seldom address residential historic areas, and methods for measuring the VCoB have yet to be refined. Therefore, taking the Xiaoxihu Historic Area in Nanjing as a case study, this work aimed to develop a more refined method for measuring the VCoB in residential historic areas based on objective physical features. The method is mainly based on the human-level perspective, identifying three key visual elements within this perspective as the objects of measurement. It collected visual data through photography, with deep learning and computer graphics processing tools being utilized to identify and extract the visual elements. Then, the study established a corresponding framework of indicators for different visual elements and optimizes the methods for indicator calculation. Through an assessment involving professionals, we validated the high accuracy of the measurement method proposed in this study. Furthermore, the study discusses factors affecting the VCoB, methods to enhance the VCoB, and the required degree of the VCoB based on the results of the measurement. The method developed in this research will provide support for the visual analysis of the urban built environment and urban renewal practices.

1. Introduction

The concept of urban renewal originated from the large-scale urban relocation and reconstruction in Western countries after World War II [1]. It has become a key component of urban policies in many countries [2]. With the accelerated global urbanization process, many cities face challenges such as urban sprawl and central area decline. Urban renewal has emerged as a significant means to counteract urban decline, gradually replacing the traditional incremental urban development model [3]. Heritage protection is a crucial aspect of this effort [4]. In its early stages, heritage protection focused primarily on “monuments”, but its scope has since been expanded to include sites, areas, cities, etc. [5]. Internationally, France was the first country to establish heritage conservation districts, passing “Legislation on the Protection of the Historical and Aesthetic Heritage of France and to Facilitate the Restoration of Real Estate” (Législation sur la protection du patrimoine historique et esthétique de la France et tendant à faciliter la restauration immobilière) in 1962 [6]. Later, in the 1960s, the U.S. National Historic Preservation Act introduced the concept of “historic districts”, and the U.K. Civic Amenities Act brought the concept of heritage conservation areas into planning law [7]. The concept of regional conservation was also adopted and developed in other countries, like the Netherlands and Australia [7,8].

UNESCO issued a “Recommendation concerning the Safeguarding and Contemporary Role of Historic Areas” in 1976, which defined “historic and architectural (including vernacular) areas as any groups of buildings, structures and open spaces including archaeological and palaeontological sites, constituting human settlements in an urban or rural environment, the cohesion and value of which, from the archaeological, architectural, prehistoric, historic, aesthetic or sociocultural point of view are recognized” [9]. In addition, residential historic areas are historic areas primarily focused on residential functions [10]. At present, research on the preservation and development of such areas encompasses historic feature protection [11,12], the economy [13,14,15], sustainability [16,17,18], etc., along with a number of comprehensive discussions [19,20].

In China, residential historic areas are common, mostly located in the central areas of cities that have a rich history and culture, with complete living facilities, convenient transportation services, and obvious location advantages regarding the surrounding areas. However, they are often shantytowns with insufficient infrastructure, dilapidated houses, and poor living conditions [21]. On the one hand, this issue arises from the increasing density of the residential population. On the other hand, it is attributed to the complex property rights problem: China’s land system has transitioned from private ownership to public ownership and then undergone a reform of socialist private housing. As a result, many families share the same plot or building, with the plot or building divided according to varying property rights (e.g., a combination of public and private ownership). This leads to unclear responsibilities in many areas, resulting in insufficient motivation or funding for building maintenance [22,23]. In the past decade, urban renewal has been an important theme in China’s urban construction and development, and residential historic areas have become one of the main objects of urban renewal efforts. The overall living environment needs to be improved, and regional vitality is waiting to be enhanced.

Vision is often the most important component of human perception [24,25,26]; humans obtain 83% of their perceptual information through vision, as compared to the other senses [27]. Therefore, study of the visual landscapes of residential historic areas plays a significant role in urban renewal efforts. Systematic analysis and research on visual landscapes commenced in the 1960s [28], covering woodlands, agriculture, coastlines, and the urban built environment. S. Kaplan and R. Kaplan developed a preference matrix that includes four visual factors: “coherence”, “complexity”, “legibility”, and “mystery”. They pointed out that “Coherence, in our perspective, helps in providing a sense of order and in directing attention. A coherent scene is orderly; it hangs together” [29]. Tevit summarized nine key visual concepts and defined coherence as a reflection of the unity of a scene, where coherence may be enhanced through repeating patterns of color and texture [30]. Bell pointed out that landscape coherence is a type of orderly structure that we can understand [31]. Sevenant and Antrop referred to coherence as the unity of a scene [32]; Kuper considered that coherence implies orderly organization [33]. Overall, visual coherence is associated with concepts of unity, uniformity, harmony, and order, making it one of the most important concepts in the analysis of visual landscapes [30]. To facilitate a more precise analysis, in this study, we define visual coherence as the degree of closeness among similar visual elements—for example, the closer colors are, the higher their coherence.

Residential historic areas are mainly composed of traditional residential buildings. Due to limitations in ancient construction techniques and the influence of the traditional culture, traditional residences often share similar visual characteristics, such as color, material, volume, height, and style, presenting an identity with a sense of wholeness. Therefore, it can be considered that such building clusters possess a strong visual coherence, which is an inherent characteristic of historic areas [34]. Coherence can help to present united and orderly visual features, enhancing the legibility of the landscape [35], guiding the observer’s attention, and helping people to understand the environment [29]. Similarity, analogy, or harmony of surface, form, or use (as in a common building materials, a repetitive pattern of bay windows, similarity of market activity, and the use of common signs) can facilitate the perception of a complex physical reality as one or as interrelated, the qualities that suggest being part of a single identity [36]. However, following industrial development, new materials and construction technologies have led to the emergence of many heterogeneous buildings, disrupting the original visual coherence of residential historic areas. As a result, urban renewal efforts need to focus on enhancing the visual coherence of residential historic areas to improve the visual quality of the landscape and protect its historic features. Moreover, before design and planning practices, it is essential to accurately measure the existing state of the visual coherence of buildings (VCoB). Based on the previous definition of visual coherence, in this paper, the VCoB can be understood as the degree of similarity in the visual characteristics within various building elements. It is associated with various building elements, such as color, material, form, size, and style, as well as their combinations [24,37].

The data sources for measuring visual landscapes primarily include images and digital models obtained from surveying and mapping [38]. In recent years, researchers have also started to explore the employment of point cloud data [39,40]. Existing research on the measurement of visual coherence typically falls into two main categories [30,41]. The first category involves subjective assessment methods, which are strongly related to observers’ experiences. Particularly in early studies, panels of judges were organized to evaluate various visual factors across different scenes, with coherence frequently being one of these factors [42,43,44,45]. However, after the 1990s, some researchers pointed out the presence of discrepancies in aesthetic evaluations across different groups [46]. They argued that relying solely on subjective ratings does not provide sufficient reliability, and the evaluation results of this method are not easily replicable [47,48]. In response, scholars gradually developed objective methods and frameworks, focusing on the inherent physical characteristics of the environment. Currently, discussions about such measurements predominantly revolve around complexity and diversity. Stamps proposed using Shannon entropy values to measure the diversity of different elements within scenes [49,50]; this indicator has also been employed by researchers to assess the complexity of plant species [33] and architectural facades [51], etc. Ewing and Handy, Wu et al., and Qiu et al. measured complexity through the quantity or proportion of elements such as buildings, vegetation, and street furniture in urban scenes [52,53,54]; Jin et al. measured the complexity of urban silhouettes through fractal dimensions [55]. In contrast, current research on objective measurement methods for visual coherence mainly targets natural landscapes. Ode et al. proposed to use the proportion of water and repeated land patterns in certain areas as indicators for measuring landscape coherence [56]; this was then applied and developed in subsequent research [57]. However, studies discussing the visual coherence of cities and buildings are less frequent. The existing relevant research primarily discusses color elements. For instance, Chen et al. and Li et al. calculated the variance in the color values within the RGB and CIELab systems, respectively, to indicate the coherence of building colors [58,59]. However, the RGB system does not closely match human perception, and the hue in the CIELab system has a unique circular characteristic, making simple variance calculations insufficiently accurate when measuring color coherence. On the other hand, color is only one of many visual elements, and focusing solely on color cannot provide a comprehensive measurement.

Therefore, we took the Xiaoxihu Historic Area in Nanjing, China, (which is one of Nanjing’s most significant residential historic areas), as a case study, aiming to develop a method that can comprehensively and accurately measure the VCoB in residential historic areas. This study addresses the gaps in the existing research by (1) analyzing the perspective of observation and the corresponding main building visual elements; (2) collecting visual data through surveying and photography, and using digital tools to identify and extract the building visual elements; (3) establishing a set of indicators to measure the coherence for different visual elements, developing and optimizing specific indicator calculation methods, and presenting the measurement results with intuitive illustrations. We organized a professional evaluation to validate the accuracy of this measurement method. Moreover, based on the results of this study, we further discuss factors affecting the VCoB, methods to improve the VCoB, and the required degree of the VCoB in residential historic areas. This research can support the analysis of visual urban landscapes and further assist related urban renewal practices.

2. Study Area

The Xiaoxihu Historic Area, located in the southeast of the old city of Nanjing, covers an area of 4.69 hm² (Figure 1). It is one of Nanjing’s most significant residential historic areas. Beyond its historic buildings and courtyards, the area serves as a vessel for traditional Chinese intangible cultural heritage, such as paper-cutting and lantern artistry, and it is associated with many historical tales and legends [60]. The Xiaoxihu Area had been established by the 10th century. After the 1950s, due to complex historical factors, the area’s population became increasingly dense, and the original property plots underwent continual subdivision. As resources shifted towards the new urban areas, development in the Xiaoxihu Area slowed, leading to a deteriorating environment [61]. A significant portion of the traditional residential buildings were destroyed, compromising the area’s historically coherent quality. In 2015, the design team from the School of Architecture at Southeast University took the lead in initiating the renewal of the Xiaoxihu Area, aiming to improve the overall living environment, while preserving historic features and introducing diverse business forms to invigorate the area. The first phase of renewal has already been completed. At present, Xiaoxihu is not only a residential area but has also become a popular sightseeing destination. Given its many transformations, the current visual features of Xiaoxihu’s buildings need to be analyzed and discussed. This will lay the groundwork for further enhancements to the visual environment in the subsequent work on the area’s renewal.

3. Materials and Methods

The research framework for measuring the VCoB comprised three primary steps. First, we defined the perspective for observing the Xiaoxihu Area and identified the main visual elements perceived in the perspective. The second step involved collecting materials of the Xiaoxihu Area’s visual landscape through on-site surveying and photography, employing deep learning and computer graphics processing tools to identify and extract the visual elements of the buildings. Lastly, we developed a specific coherence measurement method for various building visual elements (Figure 2). The following provides a detailed explanation of each step of the framework.

3.1. The Perspective and Visual Elements of Measurement

A landscape can be approached from both egocentric (i.e., human view) and exocentric (i.e., bird’s eye view) perspectives [62]. However, regarding the Xiaoxihu Area, there are no suitable vantage points nearby that can provide an aerial view. In most cases, people mainly observe Xiaoxihu by walking through the streets. Therefore, this study is primarily based on the human-level perspective.

Color is one of the elements that is most frequently discussed in research on visual landscapes. In the 1940s, Donald pointed out that “color has played a prominent part in the architecture of many lands and periods” [63]. Subsequent related research took color as a primary visual element in buildings and conducted extensive discussions [64,65,66,67]. Stamps proposed that scale and character are two important visual elements of buildings, with scale being related to the number of stories, height, and zoning density, and character being related to the building style [68]. From the human-level perspective, building overall color, height, and style are the main visual elements that people can perceive. This study thus identified these three visual elements as the focal points for measuring the VCoB. It should be clarified that there are many elements associated with the VCoB; this study only selected the main elements that are most frequently discussed.

3.2. Material Collection and Processing

3.2.1. Photography

This study required photographic documentation of the Xiaoxihu Area to gather data on the building colors, heights, and styles. Due to this area’s small scale and narrow streets, no street view data are available, necessitating on-site photography. Selecting the appropriate time for photography is crucial, as it significantly influences the visual perception of the urban landscape [69]. The photography for this study was conducted in August, a period of peak tourist activity in the Xiaoxihu Area. Additionally, to avoid the strong contrast between light and shadow caused by direct sunlight, which can affect images’ color representation, the photography was scheduled on cloudy mornings. The photography location was within the Xiaoxihu Area, and the photographers were the authors.

In the 1960s, Cullen developed the theory of “Serial Vision”, drawing a series of street scenes at predetermined angles and directions along viewing routes to present the visual sequences of streets [70]. Currently, many studies predefine and fix observation directions and angles on a street to collect street view images according to people’s observation habits [54,69,71,72]. Based on this, and considering the planning intentions of the renewal team and the preferences of most visitors, this study identified four main viewing and touring routes for the area, along with the corresponding viewing directions (since the northeastern area of Xiaoxihu is still under construction and closed to the public, we did not collect images there). Other secondary streets primarily serve as pathways for local residents to access their houses and are not highly public, and most other citizens rarely walk on these streets. Therefore, they were not the main subjects of this study. We kept the camera level and facing forward, conducting video recording along the centerline and the direction of the predetermined route. To minimize camera shaking and changes, this study utilized a DJI Pocket 2 PTZ camera fixed on a pushcart and maintained at a height of 1.6 m. After completing the recording, scene points and corresponding images were uniformly selected at 10 m intervals along the identified routes, totaling 65 scenes (Figure 3). Hence, we ultimately obtained 65 street view images.

3.2.2. Image Processing

Since this study primarily focused on buildings, it was necessary to extract them from the images. Over the past decade, artificial intelligence technology has rapidly evolved, and the study and analysis of urban images using deep learning has become an important trend [64]. Currently, semantic segmentation models such as U-net, Segnet, and Deeplab V3+ can be combined with different datasets to efficiently and extensively identify and extract various elements in urban landscapes [73,74]. They can provide significant technical support for measuring visual elements in the urban environment. Due to the relatively small number of images to be processed, developing a specific dataset and training models exclusively for this study would not be cost-effective. Therefore, we chose appropriate existing models and datasets based on available resources.

In this study, we employed the relatively advanced DeepLab V3+ model to automatically extract the building elements from various images. The DeepLab series represents a cutting-edge approach in semantic image segmentation, leveraging TensorFlow’s convolutional neural networks (CNNs) for detailed pixel-level image classification [74]. Moreover, the processing of different types of images requires matching datasets. For street view images, Ade20k, Cityscapes, and PASCAL VOC are the datasets currently common used [71,75]. In residential historic areas like Xiaoxihu, due to the narrow living spaces, household items often encroach onto public streets, adding complexity to the visual landscape. The Ade20k dataset contains different scenes from indoors and outdoors, annotating not only buildings and vehicles but also interior furniture [76]. Therefore, we used it to extract buildings from all street view images (Figure 4). The spatial constraints of Xiaoxihu and the diverse elements present in some scenes reduce the precision of the aforementioned semantic segmentation methods in specific contexts. Due to the manageable volume of image processing, we integrated manual corrections with automated methods to enhance the accuracy of building element extraction in certain challenging scenes.

3.3. Measuring the VCoB

3.3.1. Color

Color analysis was based on the buildings extracted from the images. The direct measurement of the overall color coherence of buildings in street view images might not yield accurate results, because different parts of a building naturally exhibit variations in color, due to differences in materials and construction techniques. For example, traditional residences in China’s Jiangnan region often feature the characteristics of white walls, gray tiles, and wooden doors and window frames; thus, measuring white, dark gray, and wood colors without differentiation would not result in high coherence. However, because residence construction elements, such as walls, roofs, and components and decorations (such as doors and window frames) maintain similar colors within their respective categories, they can still convey a sense of high coherence. Therefore, this study continued to divide the buildings extracted from the street view images into three parts: walls, roofs, and components and decorations. It then measured their colors separately. Most existing semantic segmentation datasets still cannot accurately identify the different parts of buildings in narrow historic area scenes. Considering the limited benefits of the specific deep learning training for Xiaoxihu, we opted for the segment anything model (SAM) to process the images. SAM is a semantic segmentation model launched by Meta AI in 2023; it leverages a large dataset, SA-1B, containing over 1.1 billion masks from around 11 million images. This comprehensive training process enables SAM to perform zero-shot generalization, allowing it to segment new images without additional training [77]. Our test showed that this tool could effectively segment the different parts of the buildings in Xiaoxihu, while some complex objects still required manual correction.

Early methods of color recognition mainly included the use of color charts, colorimeters, or spectrophotometers, etc. [78]. In recent years, with the rapid development of digital image processing technology, software libraries such as OpenCV (4.9.0) have shown strong image feature detection and description capabilities, making computer measurement and processing of images increasingly popular. Therefore, this study utilized OpenCV (4.9.0) for color recognition and analysis.

Due to the multitude of pixels and complex colors in each image, it was necessary to cluster the colors to present the main color composition of the objects. According to the conclusions of existing research, setting the number of color clusters to 4 was appropriate [64,78]. Hue, saturation, and value (HSV) constitute a color representation system that aligns more closely with human perception. The hue and saturation form a circular plane, where the hue represents the angle on the circle, ranging from 0 to 360. The saturation is the radius from the center of the circle, while value is the dimension perpendicular to the HS plane. To facilitate the calculations in computers, we set the ranges for both the saturation and value to 0–255. Together, they form a cylindrical space where all colors can be considered as points within this space. This study employed the K-means clustering method to categorize the images’ colors based on their distribution in the HSV space; it then measured the proportion of each color and generated pie charts to illustrate the dominant color composition of each object (Figure 5).

Existing research primarily presents color coherence through the variance or average color distance based on the RGB color system. The principles of both indicators are similar: the RGB color system can be considered a cubic space, and the overall color variation is measured by calculating the average value of the distances between each point (representing the color of each basic unit in the scene) and the mean point [40,58]. However, compared to the RGB system, the HSV color system is closer to human perception [64,78]. Therefore, based on the calculation of the average color distance in the RGB system, this study translated and improved it to measure the average color distance in the HSV system (which can be considered a cylindrical space). Additionally, we employed digital tools to obtain HSV values for each pixel in the image and computed the average HSV (ignoring the white background pixels of each image). The average values for the saturation and value were computed by summing the respective pixel values and dividing them by the total pixel count. However, the hue attribute exhibits circular properties; for instance, the mean hue between 60 and 300 degrees should be 0, not 180 as would result from conventional averaging methods. Therefore, the calculation of the average hue necessitates the application of Equation (1) to accurately reflect its circular nature. Through Equation (2), we can determine the average distance between each pixel’s color HSV and the average HSV, thus deriving the average color distance. The smaller the average color distance, the higher the coherence of the colors.

$$\overline{H}=\left(\frac{180}{\pi }arctan2\left(\frac{1}{n}\sum _{i=1}^{n}sin\left(\pi \frac{H_i}{180}\right),\frac{1}{n}\sum _{i=1}^{n}cos\left(\pi \frac{H_i}{180}\right)\right)\right)mod360$$

(1)

$$C{D}_{avg}=\frac{1}{n}\sum _{i=1}^n\sqrt{({\overline{S}}^2+S_i^2-2·\overline{S}·S_i·cos((H_i-\overline{H})·\frac{\pi }{180}))+{(V_i-\overline{V})}^2}$$

(2)

$\overline{H}$, $\overline{S}$, and $\overline{V}$ represent the average HSV values of the object; n denotes the total number of pixels; $H_i$, $S_i$, and $V_i$ represent the HSV values of the i-th pixel; and $C{D}_{avg}$ represents the average color distance of the object image.

Furthermore, we combined the average color distances of different parts of the buildings with their corresponding proportions in the images, to derive the whole buildings’ average color distances in each scene through Equation (3). This indicator served to present the color coherence of the whole buildings within each scene.

$$CD{B}_{avg}=CD{W}_{avg}·PW+CD{R}_{avg}·PR+CD{F}_{avg}·PF$$

(3)

$CD{B}_{avg}$ represents the overall average color distance of the buildings in each scene; $CD{W}_{avg}$ refers to the walls’ average color distance; $PW$ denotes the proportion of walls in the overall buildings of each scene; $CD{R}_{avg}$ indicates the roofs’ average color distance; $PR$ indicates the proportion of roofs in the overall buildings; $CD{F}_{avg}$ is the average color distance of the components and decorations; and $PF$ represents the proportion of components and decorations in the overall buildings.

3.3.2. Height

Semantic segmentation models and corresponding datasets capable of identifying building heights and styles in urban scenes have been developed [79,80], but their accuracy remains limited when they are applied to China’s residential historic areas, such as Xiaoxihu. Similarly, considering the limited benefits of targeted deep learning training in small-scale areas, we temporarily identified and extracted buildings with different heights and styles (which will be discussed in the subsequent subsection) through manual methods (Figure 6). For simplicity, the building heights in Xiaoxihu were measured by their number of floors, resulting in five distinct height categories, ranging from one to five floors. After categorizing the buildings by floor count in each scene, this study calculated and illustrated their respective proportions within the total buildings. Variance and standard deviation are often used to measure the degree of height variation [81]. Thus, we subsequently applied Equation (4) to calculate the average number of floors and derived the standard deviation of the building floors using Equation (5). The smaller the value, the higher the coherence of the building heights.

$$\overline{F}=\sum _{i=1}^n(F_i·P{F}_i)$$

(4)

$$S{D}_H=\sqrt{\sum _{i=1}^n({(F_i-\overline{F})}^2·P{F}_i)}$$

(5)

$\overline{F}$ is the average number of building floors; $F_i$ represents the i-th category of building floors; $P{F}_i$ is the proportion of the i-th building in the total buildings in each scene; $S{D}_H$ is the standard deviation of the building height; and n is the total number of buildings in each scene.

3.3.3. Style

Building style is a very complex and important issue; due to space limitations, it is difficult to discuss it thoroughly in this work. Additionally, since the focus of this study is primarily on the methods of measuring coherence, we simplified the issue of style, concentrating mainly on visual elements and decorations related to style. Hence, through investigation and analysis of the current situation, this study categorized the building styles in Xiaoxihu into three main types (Figure 7):

(1)

Modern style. Characterized by construction under the modern system, these buildings typically feature simple exteriors without historic symbolic decorations.

(2)

Traditional style. This category encompasses buildings constructed with historic features according to traditional construction systems. It is subdivided into the following:

• Ming and Qing (M&Q) style. This primarily refers to the building style that formed during the Ming and Qing dynasties (from the mid-14th century to the early 20th century). In the Xiaoxihu Area, buildings of this style represent significant historic heritage, necessitating restoration or reconstruction to their original historic condition. Their characteristics include dark grey brick walls (partially with white mortar finishes) and dark grey tiled sloping roofs.
• Republic of China (ROC) style. This style developed from the 1910s to the 1940s as a result of the integration of Chinese and Western construction techniques at that time. Buildings of this style are also important historic resources, primarily featuring dark grey bricks and tiles;
• Self-built style. This style was developed by the residents of Xiaoxihu between the 1950s and 1970s. Local residents evolved the style based on traditional M&Q construction techniques, building with more affordable materials (such as red bricks and tiles) and simple decorations.

(3)

Modern combined with traditional style. In the process of renewing the Xiaoxihu Area, the renovation of traditional-style buildings that are not classified as significant historic heritage often incorporates modern technology and introduces new design languages, to satisfy contemporary usage needs and aesthetic preferences. Meanwhile, some modern buildings constructed after the 1970s incorporated traditional symbols during their renewal, to harmonize them with the area’s predominant M&Q style. This resulted in a blend of modern and traditional styles: M&Q + modern, ROC + modern, and self-built + modern. Buildings of the ROC + modern style, although they exist in Xiaoxihu, are not visible on the main routes identified in this study.

We identified and extracted the building styles present in each scene, and calculated and illustrated the proportions of buildings with different styles (Figure 8). Because style cannot be digitally expressed, as with color or height, we used the Shannon entropy index [82] to measure coherence based on the proportions of different building styles. The Shannon entropy was created as a measure of disorder in information and was later introduced into the study of environmental aesthetics, where it has been used to measure the diversity of visual landscapes [49]. Nguyen et al. pointed out that Shannon entropy is also applicable for measuring coherence [78], especially when dealing with elements of the same category. The smaller the value of entropy, the larger the proportion of the dominant style in the scene, indicating a higher degree of coherence (Equation (6)).

$$H_S=-\sum _{i=1}^{n}P{S}_i{log}_{2}P{S}_i$$

(6)

$H_S$ is the Shannon entropy value of the building styles in each scene; n is the total number of building styles in the scene; and $P{S}_i$ is the proportion of the i-th building style.

4. Results

We first assigned a number to each scene (for example, the first scene of Route 1 was denoted as 1-1). Subsequently, we measured the visual characteristics and coherence of the buildings in the individual scenes and the aggregate of all scenes. These were then mapped onto the master plan of the Xiaoxihu Area, accompanied by the construction of corresponding line graphs.

4.1. Color

As previously mentioned, we clustered the colors of buildings in each scene into four groups, to represent the color composition. Considering the greater complexity of the combined color composition across all scenes, we increased the cluster number to eight, to analyze the color composition more precisely. The following will elaborate on the color characteristics of the buildings in the Xiaoxihu Area.

(1)

Wall. As shown in Figure 9 and Figure 10, gray occupies a major proportion of the wall colors throughout the Xiaoxihu Area. This primarily arises from the use of gray bricks and plaster finishes, which are common in local traditional residences. In order to harmonize with these, newly constructed buildings also predominantly employ materials of similar colors. The warm gray originates from the mortar finishes of self-built style houses. Red, on the other hand, appears in a limited number of buildings with exposed red brick walls; moreover, at the intersection of Routes 2 and 3, there stands a temple with red painting. Scenes that mix temples or red-brick buildings with other gray-walled structures tended to have a $C{D}_{avg}$ exceeding 100, presenting lower color coherence (Figure 11).

(2)

Roof. On the main routes identified in this study, a portion of the sloping roofs are not visible from the street space, due to variations in the building height and orientation. In most scenes, gray dominates, encompassing both dark gray tiles and light gray steel roofs, contributing to the generally high roof color coherence. Red-tiled roofs, observable in certain scenes along Route 4, when mixed with gray tiles resulted in high $C{D}_{avg}$ values, thereby reducing the color coherence. Additionally, at the end of Route 2, the color coherence is further compromised by a house where part of the roof is covered with a green waterproof cloth (Figure 12 and Figure 13).

(3)

Components and decoration. As presented in Figure 14 and Figure 15, most buildings feature metal components with gray coatings, or wooden components with dark red coatings. However, buildings renewed for commercial purposes often adopt vibrant colors such as green or orange to increase their attractiveness. Additionally, some residences, when replacing damaged doors and windows, choose colors without consideration of matching with other houses. These practices resulted in a greater number of scenes within the Xiaoxihu Area exhibiting lower color coherence among their components and decorations, and the regularity of the coherence variations was not strong.

(4)

Overall characteristics. We integrated the $C{D}_{avg}$ values of the overall buildings and individual parts into a single line graph (Figure 16). It can be observed that the overall $C{D}_{avg}$ of the buildings was roughly similar to the $C{D}_{avg}$ of the walls. This is primarily because the walls generally constituted the largest proportion in the scenes, and they had the most critical impact on the color coherence of the buildings.

Comparing the colors across different building parts, the $C{D}_{avg}$ value of roof colors was the lowest, indicating the highest coherence, and it showed the most gradual fluctuation along the routes. The wall colors followed, while the coherence of component and decoration colors was significantly lower, with greater fluctuations and no regularity.

4.2. Height

As seen in Figure 17 and Figure 18, the main scenes in the Xiaoxihu Area predominantly feature one- and two-story buildings, most of which have high coherence. Multi-story buildings exceeding two stories visible in the scenes are located at the intersection of Routes 1 and 4, on the south side of Route 2’s middle section, at the end of Route 4, and in the area north to Xiaoxihu. These areas possess larger property parcels, which can support the construction of relatively large multi-story buildings. When multi-story buildings are mixed with lower-rise houses in these scenes, significant height disparities occur, thereby reducing the coherence.

4.3. Style

We can see in Figure 19 and Figure 20 that self-built houses constitute the largest proportion of styles in the scenes along the main routes, primarily located at the ends of Routes 2 and 4. This is because these sections are on the periphery of the Xiaoxihu Area and have not been renewed, thus maintaining the original state of the self-built houses. Due to the scarcity of remaining significant historic buildings in this area, the visibility of M&Q and ROC style buildings is relatively limited, mainly located at the intersection of Routes 2 and 3. Other buildings exhibit predominantly modern or mixed (modern and historic) styles, primarily situated at the intersections and extended areas of Routes 1, 2, and 4. These areas represent the main focus of Phase 1 of the Xiaoxihu renewal project, where most buildings along these routes have undergone renewal and transformation, incorporating new design elements, resulting in the high prevalence of modern architectural features. The ends of Routes 2 and 4, containing predominantly self-built houses, exhibit the highest style coherence; meanwhile, the mid-section of Route 3, which has undergone partial renewal, presents an area of mixed styles and thus exhibits lower coherence.

4.4. Result Validation

This study employed a professional evaluation to validate the measurement results of Xiaoxihu’s VCoB. Given the expertise required in urban design and planning, we invited 15 graduate students and professors from this specialized field as assessment professionals. Although the results of the professional evaluation may not be entirely accurate, they could assist in validating the measurement outcomes of the method developed in this study. The professionals were allowed to freely view 65 street view images of Xiaoxihu. Due to the large number of street view images, they were required to select three scenes each with the highest and lowest coherence in color, height, and style. We then conducted a statistical analysis of the professional assessment results (

Table A1. The scenes selected by the professionals and the corresponding vote counts.

Voting Object	Scene No.	Vote Count
Scenes with the highest color coherence	1–4	13
	4–17	11
	1–5	11
	2–4	5
	2–5	2
	2–1	2
	2–18	1
Scenes with the lowest color coherence	1–13	8
	2–13	8
	2–14	6
	4–7	5
	4–13	4
	2–20	4
	2–15	3
	1–6	3
	3–4	2
	4–2	1
	3–1	1
Scenes with the highest height coherence	4–17	6
	3–2	5
	4–11	5
	2–21	5
	2–5	4
	4–8	4
	2–12	2
	4–9	2
	1–9	1
	3–1	2
Scenes with the lowest height coherence	3–8	10
	3–9	9
	1–5	7
	1–8	6
	2–8	6
	2–6	3
	4–7	2
	3–7	2
Scenes with the highest style coherence	All experts pointed out that Scenes 2–14, 2–17 to 2–21, and 4–13 to 4–17 possessed the highest building style coherence.
Scenes with the lowest style coherence	3–3	12
	3–4	9
	2–8	7
	3–6	6
	2–7	5
	4–7	3
	4–12	3

). By comparing the outcomes of the professional assessments and this study’s measurements, we found that the agreement for the scenes with the highest and lowest coherence in various visual elements reached 73.3% for the top five and 81.7% for the top ten. This indicated that the VCoB measurement method proposed in this study had a high level of accuracy.

5. Discussion

5.1. A Comprehensive and Refined Method for Measuring the VCoB

This study explored a method for measuring the VCoB in residential historic areas. The VCoB is associated with the presentation of historic features and the visual environmental quality in these areas [34]. Current studies on the objective measurement of visual landscape largely focus on complexity and diversity [33,49,50,55]. Some measurements of visual coherence primarily target natural landscapes, discussing water and land patterns [56]. However, there has been limited research on the VCoB in the urban environment. Some existing studies on the VCoB generally focused on color [58,59], lacking a comprehensive discussion of various visual elements.

This study was primarily based on the human-level perspective, integrating the three main visual elements of building color, height, and style to conduct a more comprehensive measurement of the VCoB. We developed an indicator framework and corresponding calculation methods. The standard deviation and Shannon entropy indexes were adapted for measuring the building height and style, while the measurement of the color coherence was specifically optimized. On the one hand, based on the concept of the average color distance, we improved the related calculation method in the HSV color system. On the other hand, instead of simply measuring the overall building color, we refined the measurements to separately consider the walls, roofs, and components and decorations, which provided more accurate results aligned with human perception. This method employs techniques such as deep learning to improve the efficiency and accuracy, enabling researchers to handle larger or more numerous areas in the future. The professional evaluation confirmed the high accuracy of this method.

5.2. Implications and Applications

The method developed in this study directly mapped the composition of building visual elements and the state of visual coherence in a residential historic area. This provides a foundation with which designers and planners can identify the locations of existing issues and improve the visual environment of the area. Besides residential historic areas, memorial historic areas such as groups of temples or palaces may also require coherence analysis. For instance, “Tokyo Urban Landscape Planning” involves requirements of coherence and harmony for the protection of the imperial palace [83]. Additionally, because coherence has universal significance in terms of enhancing the unity and legibility of the visual landscape [29,36], even non-historic areas (such as general residential districts) should take it into consideration; and our method is applicable to these areas as well.

To date, many researchers have explored methods to measure and evaluate the historic value of heritage areas [84,85,86]. Future research can combine these methods with the measurement of the VCoB, as discussed in this study, to further investigate the relationship between the VCoB and the historic value of architectural heritage.

Moreover, this method can be extended to measure other visual attributes. For instance, Cullen and Ashihara proposed the importance of variations in scene and spatial sequences [70,87]. Our method can be used to present variations in the visual element composition and the corresponding coherence values across viewing lines, providing a reference for representing and measuring the rhythm of scene sequences.

5.3. Factors Affecting the VCoB

During the M&Q dynasties, due to the limitations of construction technology and materials and the influence of China’s traditional feudal hierarchy, the buildings in Xiaoxihu, which is a residential area, mostly exhibited similar visual characteristics, presenting a high degree of coherence. Following modernization, the hierarchical system gradually diminished. During the replacement of old with new buildings in the Xiaoxihu, the advancement of modern technologies and materials provided support for the construction of large-volume, high-rise buildings and also introduced new building colors and styles (for example, the emergence of self-built and modern styles, as well as the use of red bricks and tiles). This is a significant premise underpinning the impact on the visual coherence of historic areas, and has consensus among most researchers. But it can not fully explain the current state of the VCoB in Xiaoxihu Area. From the measurement results presented earlier, we can find the following:

Firstly, changes in the function of buildings are an important influencing factor. Seeking to invigorate the area, the initial phase of the renewal efforts transformed some residential buildings along main streets into shops or restaurants. To attract tourists, businesses inevitably employed different colors and design languages. At the intersection of Routes 2 and 3, our design team reconstructed a temple based on historical information and designated it as a community center. To emphasize the temple’s status and its function as a public building, its exterior walls were painted red to distinguish it from the surrounding residences.

Secondly, the stability of different parts of a building is associated with visual coherence. According to our measurement results, the color coherence of components and decorations significantly lags behind that of walls and roofs. As the primary structural elements, walls and roofs are more stable and less prone to alterations. In contrast, components and decorations have lower stability, making them easier to damage or change. Regarding most residential buildings built before the 1990s in the Xiaoxihu Area, their doors and windows are mostly composed of wood, which is easily damaged, leading to more frequent repairs and replacements with diverse colors and styles. When buildings used for commercial operations change tenants, their components and decorations are often replaced to match the new occupants’ needs and preferences. Thus, lower stability leads to more frequent changes, thereby affecting the visual coherence.

Additionally, property parcels represent an often overlooked factor. The layout of property parcels is closely related to the granularity of an area’s texture [60]. In the traditional era, due to various constraints, the scale of individual residences was largely similar, with the parcel size mainly influencing the number of residences. However, under the modern construction system, the area of a parcel dictates the maximum volume of a building. In the Xiaoxihu Area, most parcels are relatively small, and houses rebuilt within these parcels may exhibit different styles or colors, but their volume is constrained by the parcel boundaries, and it is challenging to construct high-rise buildings on small parcels. However, larger parcels can provide the conditions necessary for constructing large-volume buildings. For instance, Block 9 was originally a large property parcel in the 1930s, composed of multiple houses and owned by an affluent family at that time. Over the decades, this parcel underwent partial division and consolidation, and by the 1980s, most of the parcel was still preserved. The government purchased this parcel, demolished the original dilapidated houses, and constructed a new set of five-story office buildings of a large volume according to current usage needs, significantly reducing the visual coherence of Xiaoxihu (Figure 21). Furthermore, due to complex historical reasons and the increasingly dense population, the property parcels in Xiaoxihu have been continually subdivided over the past few decades. New constructions on subdivided small parcels cannot match the scale characteristics of the original traditional residences, also affecting the visual coherence.

5.4. Methods to Enhance the VCoB

As analyzed previously, property parcels are a fundamental factor influencing building form. Thus, first, it is necessary to manage the coherence of parcel sizes. On the one hand, for overly fragmented parcels, we recommend appropriate consolidation to align them closely with the characteristics of historic parcels. If property owners cannot be coordinated, it is then advisable for them to undertake the joint renovation or reconstruction of their houses, enabling the buildings to approximate historic scales. On the other hand, it is also important to limit the size of parcels after consolidation, to prevent the emergence of large-volume buildings in future construction processes.

For already constructed large-volume or high-rise buildings, visual heterogeneity can be mitigated through design approaches. For instance, in terms of existing buildings that exceed three stories, visual obstruction can be implemented at strategic locations using trees or walls. Taking the early section of Route 2 as an example, although the low residential buildings on one side contrast with the tall substation on the other side, the scene exhibits a high degree of coherence. This coherence is achieved through a wall constructed along the street, which obscures pedestrians’ views to the substation (Figure 22).

Regarding styles, in future renovations, the Xiaoxihu Area will see more buildings being remodeled or rebuilt, integrating new construction systems and designs. Aside from a few historic buildings, a mix of modern and historical styles will dominate. On the one hand, it is necessary to retrofit purely modern buildings in the area by employing historic symbols, while on the other, styles unrelated to local history should not be permitted. In terms of colors, we need to establish a recommended color system. Based on this system, buildings that are not harmonious can undergo exterior painting or replacement of materials and decorations to achieve visual coherence.

5.5. The Required Degree of VCoB

Previous studies have shown coherence is often positively associated with high visual landscape values [29,33,47,88]. However, excessive coherence can also lead to monotony and boredom. Even in China’s ancient cities, amidst highly coherent residences, heterogeneous structures such as towers or temples exist, contributing to the visual variation in the urban landscape. Therefore, the appropriate degree of coherence warrants discussion.

The Nanjing Historic Preservation Plan (NHPP) categorizes historic areas into three levels. The first-level areas often possess a significant number of historic architectural heritage buildings, and their overall development aims to showcase historic characteristics to a greater extent, requiring high coherence; an example is the Laomendong Historic Area adjacent to Xiaoxihu. Meanwhile, the Xiaoxihu Historic Area is classified as a second-level area, where historic heritage is relatively less prevalent and some heterogeneous buildings have already been constructed, leading to a comparatively weaker overall historic character. Therefore, the NHPP stipulates a relatively lower coherence requirement for this area. Over the course of historical development, the non-traditional buildings in Xiaoxihu have gradually become part of the residents’ nostalgia. Hence, the design team selectively preserved some characteristics of these buildings during their renewal, resulting in variations and fluctuations in visual coherence across different scenes within Xiaoxihu in our measurement results. However, according to the results of the expert seminars organized by our research team, media reports, and public feedback, despite the differing degrees of coherence, both Laomendong and Xiaoxihu’s visual landscapes are beloved by the people.

From this, we can discern that criteria for the degree of coherence are diverse; they need to be linked with the characteristics and conditions of the area itself, and also depend on the judgment and intentions of designers and planners. At the same time, despite the reliability of the observer assessment method being questioned, the targeted evaluation of the area can provide support for the designers. This is one of our subsequent research aims. Therefore, future efforts to optimize the VCoB in the Xiaoxihu Area will be based on the consideration of various factors, rather than the pursuit of extreme coherence.

5.6. Limitations and Future Research

This study has some limitations, which need to be addressed and refined in future work. Firstly, a part of the Xiaoxihu Area is currently under construction, so the measurements in this study did not cover the entire area. Further measurement of the Xiaoxihu Area will be performed upon the completion of construction. Secondly, the street view images in this study were collected in one direction along four main paths and their corresponding viewing directions. However, in reality, people may change their viewing angles and directions while walking, such as viewing the area from the opposite direction to the main route (although this is not the primary direction). Therefore, future research may involve supplementing image collection from the opposite direction, conducting corresponding measurements, and creating another set of diagrams. The comparison of the two directions’ results may provide more meaningful information. We will also further explore street view collection methods that can adapt to changes in observation behavior. Thirdly, when dealing with street view images in residential historic areas, the accuracy of existing deep learning technology in identifying objects has limitations. At the same time, such areas often have a smaller scale and do not require processing a large volume of data, so we combined semantic segmentation with manual correction to improve the accuracy of measurement. While this method guarantees precision, there remains room for efficiency improvement. In subsequent work, we intend to optimize our algorithms and develop a more comprehensive dataset for Nanjing’s residential historic areas, so as to process the visual environment more efficiently. Lastly, this study primarily focused on buildings, which constitute a significant portion of the Xiaoxihu Area, but a complete visual environment also includes street furniture, plants, and other elements. Future research should include various additional elements of the area in the discussion. At the same time, in this study, we discussed only the visual attribute of coherence. Future studies can build on this research, to further explore measurement and presentation methods for other important visual attributes, such as rhythm, imageability, and complexity, etc.

6. Conclusions

This study took the Xiaoxihu Historic Area as a case study to introduce a method for measuring the VCoB in residential historic areas. To address the deficiencies of previous research, this method focuses mainly on the human-level perspective and takes the building color, height, and style as the main visual elements. It establishes a corresponding framework of indicators for measuring the VCoB, develops and optimizes specific calculation methods, and presents the results in intuitive illustrations. A professional evaluation verified the high accuracy of this measurement method. From the measurement results, we can find certain variations and fluctuations in the VCoB across different scenes in the Xiaoxihu Area, indicating that further work is required for optimization and improvement.

This research analyzed factors affecting the VCoB in the Xiaoxihu Area from the perspectives of architectural function, stability, and property parcels. Correspondingly, it proposed strategies and methods to enhance the VCoB, such as controlling parcel sizes, mitigating or concealing heterogeneous buildings, and adjusting styles and colors. It also points out that the appropriate degree of the VCoB should take into account a comprehensive range of factors, including the characteristics of the target area, design intentions, and evaluations by observers.

Our study is an exploration of measuring the visual landscape of a historic area; this is a contribution to a broader problem requiring more advanced research. In future work, we aim to further optimize the techniques and methods for identifying visual landscape elements and indicator calculation, involve a comprehensive consideration of various visual elements within residential historic areas, including buildings, trees, and street furniture etc., to develop a more refined framework for measuring visual landscapes.