Towards improving automatic image annotation using improvised fractal SMOTE approach

Abstract

It is very much essential for the multimedia information organization to provide accurate and scalable solutions to map low-level perceptual features to high-level semantics. Therefore automatic and efficient annotation of images is needed for rapid content based retrieval and indexing; it alleviates the disadvantage of any manual annotation. The proposed system for pattern matching and annotation from large image databases has been given based on the combination of Fractal Transform and gentle AdaBoost algorithm. This technique involves two main stages in classification phase wherein first, we make use of gentle AdaBoost algorithm as it is best suited for object detection task and also has lower computational complexity. Next, a mathematical representation is associated to the images of the database, this representation is a set of function parameters resulting from a dedicated fractal interpolation scheme, and used as an index by a retrieval algorithm. Proposed algorithm works completely in the Fractal transform parameter space of both images and patterns, to obtain performances well-matched with an interactive search. In this paper, we also try to overcome the orientation, scaling and class imbalance problem in image annotation by choosing an over sampling method for learning the classifier. Experimental results of IFSMOTE shows higher prediction quality, and performs better than the classical SVM, SMOTE and FSMOTE.