Automatic image annotation using an evolutionary algorithm (IAGA)
Shahram Bahrami, Mohammad Saniee Abadeh · 2014
Automatic image annotation (AIA) for a huge number of images is one of the most difficult challenging topics for researchers in the last two decades. For labeling images accurately, more various features containing low-level image features, textual tags of images have been extracted so far; however, not whole features give useful information for each conception. Feature selection as one of the important preprocessing methods, which contain the optimization of feature descriptor weights and the selection of an optimum subset feature descriptor, are desirable to improve the performance of image annotation by decreasing the feature dimension properly. In this paper, we try to propose an automated annotation based method to solve AIA in three separate phases, which is named Image Annotation Genetic Algorithm (IAGA). Principally, we use GA as feature selection in the first phase to solve the high dimensions problem, in the next phase we apply Multi-Label KNN algorithm to weight neighbors and generate a novel weighted matrix, and in the third phase we try to use GA to combine the results and assign the related words to new images. We employ two well-known and the most important datasets, Corel5K and IAPR TC-12. The experimental results show that the proposed method outperforms other well-known methods and can be expeditiously employed to solve the multi-model engineering problems with high dimensionality.