Algorithmic Research For Automatic Image Annotation Based On Color Constancy And Multiple Instance Learning

Automatic image annotation has been an active research topic in recent years due to its potentially fundamental impact on image understandings and that manual image annotation for indexing and then later retrieving image collections is an expensive and labor intensive procedure.In this paper, a novel automatic image annotation algorithm is proposed, which extends color constancy and Multiple Instance Learning (MIL), and their applications to the problem of block-based image annotation. Images are viewed as bags, each of which contains a number of instances corresponding to blocks obtained from image sub_blocking. To correct images using an appropriate color constancy method can make low-level features more robust. Then we can apply MIL technique to get more proper instance prototypes to construct bag features which represent corresponding categories. The thesis first introduces some classical color constancy algorithms and applies general Gray-World algorithm to preprocess images, which is based on grey-world hypothesis. A new image sub_blocking scheme is proposed, which can improve the efficiency. Then the bag features are obtained by applying MIL technique on the image blocks, where the enhanced DD algorithm and a faster searching algorithm are applied to improve the efficiency and accuracy. Finally, the bag features are input to a set of SVMs for finding the optimum hyperplanes for automatically annotate images.Our proposed annotation approach demonstrates a promising performance for an image database of 2000 general-purpose images from COREL, as compared with some current peer algorithms in the literature.