Texture Retrieval Effectiveness Improvement Using Multiple Representations Fusion

Abstract

We propose a multiple representations approach to tackle the problem of content-based image retrieval effectiveness. Multiple representations is based on the use of multiple models or representations and make them cooperate to improve search effectiveness. We consider the case of homogeneous textures. Texture is represented using two different models: the well-known autoregressive model and a perceptual model based on perceptual features such as coarseness and directionality. In the case of the perceptual model, two viewpoints are considered: perceptual features are computed on original images and on the autocovariance function corresponding to original images. Thus, we use a total of three representations (models and viewpoints) to represent texture content. Simple results fusion models are used to merge search results returned by each of the three representations. Benchmarking carried out on the well-known Brodatz database using the recall graph is presented. Retrieval relevance (effectiveness) is improved in a very appreciable way with the fused model.