Abstract
Content-Based Image Retrieval (CBIR) is a new emerging technology for image retrieval purpose. So, this paper presents an experimental analysis on document image retrieval using a multi-domain intelligent system called ISIR (Intelligent System for Image Retrieval). ISIR system investigates, on the same document image, the combination of three different domains: color, layout and texture. More specifically, the extracted features by these three domains are provided to the system for clustering and classification tasks. K-Means algorithm has been used for clustering and Euclidean Distance, Dynamic Time Warping (DTW) and Support Vector Machine (SVM) have been used, respectively, for the classification of color, layout and texture. The combination of individual decisions was considered. The experimental results, carried out on 82 forms belonging to 20 different classes, demonstrate the effectiveness and efficiency of the proposed approach respect to the literature, based on the ANR performance index and time consumed. The ANR parameter is able to evaluate the multi-domain system. Finally, all data automatically flow into a specific database for the industrialization process.
Keywords: image retrieval, intelligent system, multi-domain combination
