Co.Vi.Wo. Descriptor

S. Α. Chatzichristofis, C. Iakovidou, Y. S. Boutalis and O. Marques “COVIWO: COLOR VISUAL WORDS BASED ON NON-PREDEFINED SIZE CODEBOOKS”, «IEEE Transactions on Systems Man and Cybernetics, Part B», Accepted for publication, 2012.


Due to the rapid development of information technology and the continuously increasing number of available multimedia data, the task of retrieving information based on visual content has become a popular subject of scientific interest. Recent approaches adopt the Bag-Of-Visual-Words (BOVW) model to retrieve images in a semantic way. BOVW has shown remarkable performance in Content Based Image Retrieval (CBIR) tasks exhibit better retrieval effectiveness over global and local feature representations. The performance of the BOVW approach depends strongly, however, on predicting the ideal codebook size: a difficult and database-dependent task. The contribution of this paper is three-fold. First, it presents a new technique that uses a self- growing and self- organized GAS neural network to calculate the most appropriate size of a codebook for a given database. Second, it proposes a new soft weighting technique, whereby each local feature is classified into only one visual word with a degree of participation. Third, by combining the information derived from the method that automatically detects the number of visual words, the soft weighting method and a color information extraction method from the literature, it shapes a new descriptor, called Co.Vi.Wo. (Color Visual Words). Experimental results on two well-known benchmarking databases demonstrates that the proposed descriptor outperforms 15 contemporary descriptors and methods from the literature, both in terms of precision-at-K (P@K) as well as in its ability to retrieve the entire ground truth.

Co.Vi.Wo. Core Experiments:

From this page you can download all the different sized codebooks used in the core experiments as xml files together with the files that contain the accumulative probabilities needed for the calculation of the soft weighting technique for both databases that were used. Additionally,  all the trec-files which include the detailed ranking lists of the experiments are also available for download.

Download the codebooks (and the accumulative probabilities) for the UCID database

Download the codebooks (and the accumulative probabilities) for the NISTER database

Download the trec_files which include the detailed ranking lists of the experiments on the UCID database

Download the trec_files which include the detailed ranking lists of the experiments on the NISTER database

The implementation of Co.Vi.Wo. is freely available through the image retrieval laboratory img(Rummager), while a second, nightly build,  implementation of the method is included in the open source image retrieval engine Golden Retriever.