Revisiting HOLC redlining maps with street view image data on urban visual perceptual quality: A case study in Buffalo

Source abstract

Redlining, institutionalized by the HOLC in the 1930s, was a discriminatory credit mapping system that divided residential areas into A–D grades based on neighborhood conditions. Although long abolished, its unequal impacts persist. Previous studies mainly focused on racial and socioeconomic inequalities, while this study used street view image analysis to examine how redlining has produced inequalities in Visual Perceptual Quality through differences in the built environment. Taking Buffalo as a case, this study applied computer vision and machine learning with the Place Pulse 2.0 dataset to predict six perceptual dimensions—beauty, safety, wealth, liveliness, boring, and depressing. The results reveal significant differences in perception scores across redlining grades, indicating that inequality persists at the perceptual level. SHAP analysis further shows that these disparities are mainly driven by built environment features like trees, buildings, and roads. The uneven distribution of these key features across redlining grades forms the physical basis of perceptual inequality. The study also discussed the relationship between such inequality and the long-term planning bias and uneven investment shaped by redlining. This study expands the research scope of redlining and provides quantitative evidence for improving disadvantaged neighborhoods long affected by its legacy.