Abstract: We propose a perception-enhanced visual analysis system for constructing and exploring high-quality infographic datasets. The system consists of three modules: control view, grid view and sample view, supporting hierarchical exploration and sample filtering. The system combines convexity-based layout optimization with dynamic color assignment to enhance cluster perception and improve analytical accuracy. A hierarchical exploration workflow then guides users from overview to detail, allowing them to filter out non-target samples, interrogate design variations, and identify gaps for targeted data augmentation. Case studies show the system effectively excludes irrelevant samples, explores the design space, and augments data. Evaluation results reveal that 100% of the sampled data are target samples, and the design space covers approximately 94% of the samples, demonstrating the system's ability to help analysts better understand and build high-quality infographic datasets.