Carpet Map Classification and Retrieval based on Color and Textural Features
Mojdeh Vafaei, Hossein Pourghassem · International journal of imaging and robotics · 2015
Carpet map image classification and retrieval by computer are important issues in Persian carpet industry. In this paper, a novel carpet map classification and retrieval algorithm based on color and texture information is proposed. Firstly, the carpet map images are classified into two major classes (Kheshti map class and Lachak-Toranj map class) and then, images are retrieved. For this purpose, directional and textural features are extracted and a feature vector is created. A Support Vector Machine (SVM) as a classifier uses the weighted feature vector to define the class of the image. To evaluate the accuracy of classification, Multilayer Perceptron (MLP) is also applied. The image retrieval is performed on the images of specified class. To reduce search space and increase retrieval accuracy, the images are filtered by our proposed color and texture filters. The proposed color and texture filters pass the images that have the highest similarity to a query image in the color and texture contents, respectively. The proposed algorithm is evaluated on a set of carpet map images that had been captured from real carpets.