ABSTRACT

Video mining is used to discover and describe interesting patterns in video data, which has become one of the core problem areas of the data mining research community. Compared to the mining of other types of data (e.g., text), video mining is still in its infancy, and an under-explored field. There are many challenging research problems facing video mining. Video Association Mining is a relatively new and emerging research trend. It consists two key phases are (i) Video pre-processing and (ii) Frequent Temporal Pattern Mining. The first phase converts the original input video to a sequence format. The second phase concerns the generation of frequent patterns. Frequent pattern generation plays an essential role in mining of association rules. The usual framework is to use a minimal support threshold to obtain all frequent patterns. However, it is nontrivial for users to choose a suitable minimal support threshold. The paper addresses the issue of frequent temporal pattern mining and studies algorithms for the same. In this paper, we proposed a new mining task called mining Best-N frequent patterns, where N is the largest rank value of all frequent patterns to be mined. An efficient algorithm called Modified VidApriori is used to mining Best-N frequent patterns. During the mining process, the undesired patterns are filtered and useful patterns are selected to generate other longer potential frequent patterns. This strategy greatly reduces the search space. The existing Apriori based algorithm is compared with Modified VidApriori. We also presented results of applying these algorithms to a synthetic data set, which show the effectiveness of our algorithm.