FP-growth algorithm for mining association rules is divided into two phases: building a FP-tree and mining frequent patterns. In this paper new algorithms are proposed to improve the two stages separately. In the first stage, if frequent items in header table have the same support, their position can be dynamically changed to construct a compressed and optimized FP-tree. IMFP-tree algorithm is proposed to realize that function. In the second stage, CFP-mine algorithm is proposed, which constructs pattern-base by using a new method different from the conditional pattern-base in FP-growth.This paper mines frequent itemsets with a new combination method without recursive construction of conditional FP-tree. It has theoretically proved and experimentally verified the correctness and efficiency of CFP-mine algorithm.