Contribution for Content based Image Retrieval by Multiple Descriptors
Abdelkhalak Bahri, Hamid Zouaki, Youssef Bourass · International Journal of Multimedia and Ubiquitous Engineering · 2017
One of the current challenges in the CBIR (Content Based Image Retrieval) domain is to find the best combination of several descriptors. Recently, several studies have been led to overcome this problem by integrating the genetic programming, however this technique required an important calculation time. The high computational cost is due to the high number of function calls required by this method and the high dimension of image descriptors used.To reduce this calculation time, we present a new search method that uses a multi-resolution BOF (Bag of Features) and genetic programming.In this context, for each descriptor type (e.i.color, shape, etc.), each image is described by a pyramid of histograms (pyramid of BOF), with each histogram represents the lower resolution of its above histogram.We start the similarity search by applying the GP (Genetic programming) to the lower histograms (top level of pyramid) for each descriptor type.In addition, we repeat this process just on the result obtained but by using the histograms correspond to the next levels of each pyramid of each image to refine the result obtained.The proposed method improves the search time without deteriorates the result quality in term of precision.We validate our method against three databases.We have shown that our approach provides interesting and powerful experimental results.