Phishing Web Resource Clustering Using Visual Contour Similarity

The patent segments web resources into different clusters based on similarity analysis. By dividing the large set of suspicious web resources into smaller, more focused clusters, the system can process each cluster separately with reduced computational overhead. This segmentation allows analysts to concentrate on specific groups of similar resources rather than examining all suspicious resources individually, thereby reducing labor intensity while maintaining identification speed.

The patent performs preliminary clustering and filtering of web resources before detailed analysis. By pre-organizing web resources into clusters based on initial similarity metrics, the system prepares the data in advance so that subsequent malicious resource identification can be performed more efficiently. This preliminary action reduces the workload during the actual analysis phase by presenting pre-organized, smaller subsets of resources.

The patent extracts and removes already identified malicious web resources from the clustering process. By taking out known malicious resources and focusing clustering efforts on suspicious resources of unknown status, the system avoids redundant processing of already-analyzed items. This extraction approach maintains comprehensive identification accuracy by concentrating computational resources on the most uncertain cases while reducing overall processing time.

The patent applies partial clustering by focusing on subsets of suspicious web resources that meet specific criteria rather than clustering all suspicious resources uniformly. By applying clustering selectively to the most promising or suspicious subsets, the system achieves reliable identification of malicious resources without the excessive processing time that would result from analyzing every single suspicious resource in detail.

The patent replaces complex mechanical or manual analysis methods with automated computer vision-based contour extraction and image processing. By using algorithmic approaches to automatically extract visual features and contours from web resource screenshots, the system achieves high clustering accuracy without requiring manual intervention. This substitution of automated image processing for manual or simpler mechanical analysis methods improves precision while managing computational complexity through efficient algorithms.

The patent creates visual content copies (screenshots) of web resources and performs contour extraction on these copies rather than analyzing the original web resources directly. By working with replicated visual representations, the system can apply complex image processing techniques to improve clustering accuracy without affecting the original resources. This copying approach allows for repeated analysis and experimentation with different contour extraction parameters while maintaining the integrity of the source materials.