Mining regular/frequent itemsets is very important concept in association rule mining which shows association among the variables in huge database. the classical algorithm used for extracting regular itemsets faces two fatal deficiencies .firstly it scans the database multiple times and secondly it generates large number of irregular itemsets hence increases spatial and temporal complexties and overall decreases the efficiency of classical apriori algorithm.to overcome the limitations of classical algorithm we proposed an improved algorithm in this paper with a aim of minimizing the temporal and spatial complexities by cutting off the database scans to one by generating compressed data structure bit matrix(b_matrix)-and by reducing redundant computations for extracting regular itemsets using top down method. theoritical analysis and experimental results shows that improved algorithm is better than classical apriori algorithm.