We are developing a system that improves access to a large, growing image collection by supporting users to collaboratively build a faceted (multi-perspective) classification schema. For collections that grow in both volume and variety, a major challenge is to evolve the facet schema, and to reclassify existing objects into the modified facet schema. Centrally managed classification systems often find it difficult to adapt to evolving collections. The proposed system allows: (a) users to collaboratively build and maintain a faceted classification, (b) to systematically enrich the user-created facet schema, and (c) to automatically classify documents into an evolving, user-managed facet schema. In this paper, we focus on (c), where we describe the approach to automatically classify documents into an evolving facet schema. We propose a learning-based system that periodically learns from manually classified images, and then classify new images accordingly.