Fuzzy aesthetic semantics description and extraction for art image retrieval

More and more digitized art images are accumulated and expanded in our daily life and techniques are needed to be established on how to organize and retrieve them. Though content-based image retrieval (CBIR) made great progress, current low-level visual information based retrieval technology in CBIR does not allow users to search images by high-level semantics for art image retrieval. We propose a fuzzy approach to describe and to extract the fuzzy aesthetic semantic feature of art images. Aiming to deal with the subjectivity and vagueness of human aesthetic perception, we utilize the linguistic variable to describe the image aesthetic semantics, so it becomes possible to depict images in linguistic expression such as ‘very action’. Furthermore, we apply neural network approach to model the process of human aesthetic perception and to extract the fuzzy aesthetic semantic feature vector. The art image retrieval system based on fuzzy aesthetic semantic feature makes users more naturally search desired images by linguistic expression. We report extensive empirical studies based on a 5000-image set, and experimental results demonstrate that the proposed approach achieves excellent performance in terms of retrieval accuracy.