Image Classification in Cultural heritage رسالة ماجستير

Abstract

In this thesis, an automated supervised image classification technique specifically for classifying images in the cultural heritage domain is developed. In general, most of the image classification techniques are used for known and semi-visible models that has known objects to detect, while the developed technique classifies images according to a particular date, culture, people and historical age. The proposed technique is comprised of two stages, the first is extracting features using unsupervised segmentation technique, then the unsupervised classification stage. The developed technique uses only hue from the CIE LAB color space for segmentation and K-means is used for Clustering. Some segments are merged to get the result of the cultural heritage to which it has the most relevance for these segments. In the learning phase, common features were extracted and then compared their histograms, then they were categorized accordingly. Finally, adjacent columns in the histograms were merged to reduce the complexity of the algorithm. Finally following the technique and applying it on a repository of cultural heritage images, reduced the complexity of the algorithm and get accuracy approximately to 70% for 20 cultural heritages, each of which has about 150 images from 3200 images in whole dataset.