Mining image frequent patterns based on a frequent pattern list in image databases

• Published in: The Journal of supercomputing Vol. 76; no. 4; pp. 2597 - 2621
• Main Authors: Chang, Ye-In, Shen, Jun-Hong, Li, Chia-En, Chen, Zih-Siang, Tu, Ming-Hsuan
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2020; Springer Nature B.V
• Subjects: Candidates; Compilers; Computer Science; Data mining; Deep Learning; Image databases; Interpreters; Parallel Computing in Biomed Sciences & Healthcare; Processor Architectures; Programming Languages; Frequent image patterns; Image databases; Image mining
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-019-03041-y

The goal of image mining is to find the useful information hidden in image databases. The 9DSPA-Miner approach uses the Apriori strategy to mine the image database, where each image is represented by the 9D-SPA representation. It presents a reasoning method to reason the unknown spatial relation that satisfies the spatial consistency. However, it may generate invalid candidates with the impossible relations that cannot be found in the 2D space or in the input database. Moreover, in this approach, counting the support of the pattern needs to intersect the associated image sets by searching the index structure, taking a long time. Therefore, in this paper, we propose an approach with a frequent pattern list, which generates all valid candidates of frequent patterns. Based on the frequent pattern list, the proposed approach presents two conditions in the candidate generation for finding frequent spatial patterns to avoid generating impossible candidates. Moreover, the proposed approach uses an additional verification step to further avoid generating impossible spatial relations. Therefore, the proposed approach generates fewer candidates than the 9DSPA-Miner approach, reducing the processing time. The experimental results have verified that the proposed approach outperforms the 9DSPA-Miner approach.