Classification of textured images based on new information fusion methods

• Published in: IET image processing 2019-07, Vol.13 (9), p.1540-1549
• Main Authors: Zitouni, Abdelkader, Benkouider, Fatiha, Chouireb, Fatima, Belkheiri, Mohammed
• Format: Article
• Language: English
• Subjects: Brodatz datasets; classifier fusion method; DTD datasets; estimated feature vector; feature extraction; Gabor features; grey‐level co‐occurrence matrix; image classification; image fusion; image segmentation; image texture; information fusion methods; neural nets; neural networks classifier; Research Article; sliding window; supervised classification algorithms; synthetic images; textured image classification; textured‐images segmentation
• ISSN: 1751-9659; 1751-9667; 1751-9667
• DOI: 10.1049/iet-ipr.2018.6256

This work presents supervised classification algorithms based on information fusion for textured-images segmentation. Gabor features are efficient in finding class boundaries, whereas grey-Level co-occurrence matrix features are favourable in the areas within the classes. Moreover, the wavelets can represent textures at different scales and offer great discriminatory power between textures with strong resemblances. This motivates us to combine these three kinds of features with improving image segmentation. In the first step, the proposed method applied those three feature extraction strategies on textured images to get more information. After that in the second step, the estimated feature vector of each pixel is sent to the neural networks classifier for pre-labelling. Then, in the third step of the proposed method, a classifier fusion method used to combine the scores obtained by the neural networks for each pixel. Finally, in the last step, to obtain more precise segmentation results, the scores within a sliding window are combined. The performance of the proposed segmentation algorithms was evaluated on synthetic images from Brodatz and DTD datasets. The obtained classification results from the proposed fusions system lead to higher classification precision compared to applying a single classifier on the textured images.