Robust image classification using multi-level neural networks
Samy Sadek, Ayoub K. Al-Hamadi, Bernd Michaelis, Usama Sayed · 2009
Image classification problem is one of the most challenges of computer vision. In this paper, a robust image classification approach using multilevel neural networks is proposed. In this approach, each image is fixedly divided into five regions each equal to half of the original image. Then these regions are classified by the multilevel neural classifier into five categories, i.e., ¿sky¿, ¿water¿, ¿grass¿, ¿soil¿ and ¿urban¿. Both color moments and multilevel wavelets decomposition technique are used to extract features from the regions. Such features have been experimentally proved to be computationally efficient and effective in representing image contents. Experimental results clarify that the proposed approach performs better than other state-of-the-art classification approaches.