A novel algorithm for associative classification of image blocks

Because of its accurate and robust performance, association rule-based approach is recently used for image classification. However, the existing algorithms for associative classification suffer from inefficiency. Addressing this problem, a novel algorithm based on atomic association rules is presented and successfully used in image block classification. Mining only the atomic association rules achieves fast image block classification. Using the strong atomic association rules, extracted under a high confidence threshold, can accurately differentiate instances from the image dataset. Furthermore, multi-passes of partial classifications can classify the whole dataset. This algorithm uses a self-adaptive confidence threshold and a dynamic support threshold, both of which are important for good classification performance. The experiments were performed on a standard dataset of image segmentation. The results show the proposed algorithm can classify the image blocks faster, more accurate and robust than the typical associative classification algorithm.