Document Image Retrieval Based on Texture Features and Similarity Fusion

In this paper we investigate the usefulness of two different texture features along with classification fusion for document image retrieval. A local binary texture method, as a statistical approach, and a wavelet analysis technique, as a transform-based approach, are used for feature extraction and two feature vectors are obtained for every document image. The similarity distances between each of the two feature vectors extracted for a given query and the feature vectors extracted from the document images in the training step are computed separately. In order to use the properties of both features, a classifier fusion technique is then employed using a weighted average fusion of distance measures obtained in relation to each feature vector. The document images are finally ranked based on the greatest visual similarity to the query obtained from the fusion similarity measures. The Media Team Document Database, which provides a great variety of page layouts and contents, is considered for evaluating the proposed method. The results obtained from the experiments demonstrate a correct document retrieval of 65.4% and 91.8% in the Top-1 and Top-10 ranked document list, respectively.