Combining structural descriptions and image-based representations for image, object, and scene recognition

摘要

Object and scene learning and recognition is a major issue in computer vision, in robotics and in cognitive sciences. This paper presents the principles and results of an approach which extracts structured view-based representations for multi-purpose recognition. The structures are hierarchical and distributed and provide for generalization and categorization. A tracking process enables to bind views over time and to link consecutive views. Scenes can also be recognized using objects as components. Illustrative results are presented. 1