Mining frequent patterns has been studied popularly in data mining research. For getting the real useful frequent patterns, one must continually adjust a minimum support threshold. Costly and repeated database scans were done due to not maintaining the frequent patterns discovered. In this paper, we first propose a top-down algorithm for mining frequent patterns, and then present a hybrid algorithm which takes top-down and bottom-up strategies for incremental maintenance of frequent patterns. Efficiency is achieved with the following techniques: large database is compressed into a highly condensed and dynamic frequent pattern tree structure, which avoids repeated database scans, the top-down mining approach adopts a depth first method to avoid the recursive construction and materialization of conditional frequent pattern trees, which dramatically reduces the mining cost. The performance study shows that our algorithm is efficient and scalable for mining frequent patterns, and is an order of magnitude faster than FP_growth and Re-mining.