In this paper we propose an approach for mining association rules in large, dense databases. For finding such rules, frequent itemsets must first be discovered. As finding all the frequent itemsets is very time-consuming for dense databases, we propose an algorithm that is able to quickly discover an image of the complete set containing all the frequent itemsets. We define what an image is, and we present a genetic algorithm for discovering such an image. To monitor the discovery process we introduce the notion of dynamics of the algorithm. To measure the performances of our frequent itemsets discovery algorithm, we introduce the notion of efficiency of the discovery process.