Comparative Study of Semantic Mapping of Images

Semantic mapping is typically used to find deep connections between different words or text documents. The purpose of this study is to extend the semantic mapping method to images. For this purpose, the publicly available affective image database was used. Five methods of semantic map construction were compared, including the methods based on (a) human ranking, (b) fMRI data, (c) metadata, (d) predominant colors found in images, and (e) analysis of verbal description of images. The compatibility of outcomes was evaluated using traditional and canonical correlation. Overall, results obtained using different methods appear to be compatible with each other, except that the correlation of the predominant color and semantic measures was found not significant. Comparative strengths and weaknesses of these approaches are discussed. The obtained semantic map provides an insight into the interpretation of brain activity recorded via fMRI, and will be useful for building and validating models of human emotional cognition.