Use of relevant indicators of Correspondence Analysis to improve image retrieval

We are concerned by the use of Factorial Correspondence Analysis (FCA) for image retrieval. FCA is designed for analyzing contingency tables. For adapting FCA on images, we first define "visual words" computed from Scalable Invariant Feature Transform (SIFT) descriptors in images and use them for image quantization. At this step, we can build a contingency table crossing "visual words" as terms/words and images as documents. The method was tested on the Caltech 4 and Stewénius and Nistér datasets on which it provides better results (quality of results and execution time) than classical methods as tf*idf or Probabilistic Latent Semantic Analysis (PLSA). To scale up and improve the retrieval quality, we propose a new retrieval schema using inverted files based on the relevant indicators of Correspondence Analysis (quality of representation and contribution to inertia). The numerical experiments show that our algorithm performs faster than the exhaustive method without losing precision.