Image Annotation Based on Feature Weight Selection

Multimedia content description interface (MPEG-7) includes a number of image feature descriptors to represent low-level image features such as colors, textures and shapes effectively. But, the contribution of each descriptor may not be the same for a domain specific image database when computing the similarity measure. Machine learning techniques for the optimization of feature descriptor weights are desirable to enhance the accuracy of image annotation systems. In our system, we use a real coded chromosome genetic algorithm and support vector machine (SVM) classification accuracy as fitness function to optimize the weights of MPEG-7 image feature descriptors. The experimental results over 2000 classified Corel images show that with the real coded genetic algorithm, the accuracies of image annotation system are improved comparing to the method without machine learning techniques.