Indoor openness offers a critical reference to how users perceive an interior space. Past studies have accumulated a diverse range of impacting factors based on visual evidence and empirical data through simulated simplistic environments, yet have failed to associate such evidence with complex real-world spaces to provide evaluations. This study introduces a framework that combines computer vision and machine learning to quantitatively assess perceptions of openness in office spaces based on digitally generated visual data, feature extraction, and perception rating data. Machine learning models are trained to predict openness ratings based on the proportions of different elements. Our results demonstrate a method to integrate multiple elements for evaluating the perception of openness, and highlight that collective understanding of the concept of openness influences users’ perception.