Image Indexing and Retrieval with Pachinko Allocation Model: Application on Local and Global Features

Abstract

AbstractWe present in this paper a part of our work in the field of image indexing and retrieval. In this work, we are using a statistical probabilistic model called Pachinko Allocation Model (PAM). Pachinko Allocation Model (PAM) is a probabilistic topic model which uses a Discrete Acyclic Graph (DAG) structure to present and learn possibly correlations of topics which were responsible of generating words in documents, like other topic models such as Latent Dirichlet Allocation (LDA), PAM was originally proposed for text processing, it can be applied for image retrieval since we can assume that image is a text and parts of image (local points, regions ,…) can represent visual words like in text processing field. We propose to apply PAM on local features extracted from images using Difference of Gaussian and Salient Invariant Feature Transform (DoG/SIFT) techniques. In a second part, PAM is applying on global features (color, texture …), these features are calculated for a set of regions resulting from 4×4 division of images. The proposition is under experimental evaluation.KeywordsImage Indexing and RetrievalPachinko Allocation ModelScale Invariant Feature TransformImage Local FeaturesColor Texture FeaturesGlobal Features