To explore whether the Latent Dirichlet Allocation (LDA) topic model is suitable for analyzing the short text of tourism microblogs and whether the result can be consistent with the research results based on interviews and other data, this study established a destination image perception framework including four first-level dimensions and 10 second-level dimensions. Then the meaning of the topics is defined based on the dominant dimensions of each topic, which can reduce the subjectivity of researchers and help them to use the LDA model to extract the destination image perception quantitatively and objectively. The case study of the Old Town of Lijiang shows that in the first-level dimensions, the basic framework of image perception can be fully outlined through the five groups of core spatial and landscape elements, including the human settlements, music culture, character, leisure space and Naxi cuisine, and the special perception of the deep tourists, urban residents, young people and girls, and the characteristics of interaction of human and environmental elements. In the second-level dimensions, more detailed perception of destination image can be vividly presented from three aspects: the slow living in the Old Town of Lijiang, the culture of nightlife and romance, and tourists' perception of the fusion of local culture and modernity. This study proves the feasibility and advantage of this method—it shows that LDA is suitable for short text analysis of social media such as Weibo. Topic analysis based on dominant semantic dimensions successfully portrays the image perception of the Old Town of Lijiang and further analyzes the mechanism of image formation, and provides a new perspective for destination image perception, which has three values. It helps to accurately establish the basic framework of destination image perception; quantitatively extract the core dimensions of image perception; and deeply interpret the local meaning of destination image and clarify the relationship between cognitive, affective, and behavioral images.