关键词: 图像检索, 词袋模型, 最长公共视觉词串

Abstract: The model of Bag-of-Features（BoF） has been widely used in image retrieval. In the BoF model, each image is represented as a frequency histogram of visual words in a codebook. It ignores the spatial context among the visual words, which is important for image representation. This paper proposes a novel image retrieval method based on the longest common visual substring. The substring is extracted based on the topology between the visual words, which contains lots of image spatial information. The experimental results on Holiday image database demonstrate that the proposed method for image retrieval outperforms the original BoF model and its improved Eo-BoW model.

Key words: image retrieval, bag-of-feature, longest common visual substring