Improving the Performance of Color-Based Signatures through Dynamic Selection of Adequate CCV-Threshold
Mawloud Mosbah, Bachir Boucheham · 2016
Color Coherence Vector (CCV) is an indexing method derived from histograms signature that attempts to integrate spatial information in these structures. Owing to its remarkable importance, many recent works have addressed CCV signature and some early indexing methods have been inspired from it. An essential issue with CCV is how to set the threshold value allowing distinction between coherent and non-coherent regions. Usually, this threshold, whose value has dramatic impact on CBIR performances, is set statically by the developer of the CBIR system. In this paper, we focus on the CCV threshold value through proposing two algorithms setting it dynamically according to the image query. The first algorithm belongs to the soft computing and relies on features fusion mechanism. The second one belongs to the cloud computing paradigm and relies on Google Image Engine. The two proposed algorithms belong to the pseudo relevance feedback approach that does not require any user assistance. The experiments conducted on the Wang database (COREL-1K) reveals that the proposed algorithms yield very promising results where they enhance CCV performances with fixed threshold and they are efficient compared to the features Fusion mechanism when considering the indexing time.