Developing an urban streetscape indexing based on visual complexity and self-organizing map

Abstract

Streetscape examinations in the digital context offer a wealth of geospatial data and application support for urban informatization, facilitating a more scientific and efficient comprehension of the city image. Currently, digital investigations on streetscapes predominantly emphasize object-based parsing rather than perception-based parsing. Furthermore, there is a notable absence of a comprehensive analytical framework specifically designed to urban visual environments. Consequently, the accurate recognition and effective management of the city image have been limited. Therefore, this study parses streetscapes from the perspective of their visual perception, and accordingly develops a digital urban streetscape indexing to analyze urban visual environments. Specifically, the streetscape is decoded into multi-characteristic visual complexity including texture, shape, and color, which derive a three-dimensional dataset. The dataset is then fed into a machine learning technique, a self-organizing map (SOM), for synthetic training, resulting in an indexing that sheds light on the interconnections between the visual characteristics, the streetscape, and its geo-distribution, thereby enabling a multifaceted analysis of the urban visual environment. Three relevant applications of the proposed indexing are subsequently demonstrated. This study indicates that the streetscape can be parsed by multi-characteristic visual complexity of texture, shape, and color; based on this, the developed indexing can function as a digital system that facilitates streetscape management and exploration. The theoretical and technical contributions of this study can support the sustainable development of city image within the digital context.