II-20: Intelligent and pragmatic analytic categorization of image collections.

Area of Science:

• Multimedia Analytics
• Human-Computer Interaction
• Computer Vision

Background:

• Existing image collection visualizations lack tight integration with machine models for analytic categorization.
• Current computer vision and interactive learning techniques often prioritize search over adaptive categorization.
• Analytic categorization differs from machine classification due to its dynamic, human-driven nature, highlighting the 'pragmatic gap'.

Purpose of the Study:

• Introduce 11-20 (Image Insight 2020), a multimedia analytics approach to bridge the pragmatic gap in image collection categorization.
• Develop a flexible machine model that dynamically adapts to user-defined categories of relevance.
• Enhance semantic interactions and user insight through novel visualization and interaction metaphors.

Main Methods:

• Developed a novel machine model that dynamically models user relevance categories and supports sliding on the exploration-search axis.
• Introduced a Tetris metaphor for 1-image-at-a-time analysis, complementing traditional grid metaphors.
• Implemented a 'fast-forward' interaction for rapid expansion of relevance categories.

Main Results:

• The 11-20 machine model matches and exceeds state-of-the-art performance in generating relevant suggestions.
• Automated experiments demonstrate the superior analytic quality of the Tetris metaphor.
• User studies confirm 11-20 as an intuitive, efficient, and effective multimedia analytics tool.

Conclusions:

• 11-20 effectively closes the pragmatic gap in analytic categorization of image collections.
• The dynamic machine model and novel interaction metaphors significantly enhance user insight discovery.
• 11-20 represents a significant advancement in multimedia analytics for image collections.