View-Based Recognition Using an Eigenspace Approximation to the Hausdorff Measure

View based recognition methods such as those using eigenspace techniques have been successful for a number of recognition tasks Currently however such approaches are relatively limited in their ability to recognize objects which are partly hidden from view or occur against cluttered backgrounds In order to address these limitations we have developed a new view matching technique based on an eigenspace approximation to the generalized Hausdor measure This method achieves the compact storage and fast indexing that are the main advantages of previous eigenspace view matching techniques while also being tolerant of partial occlusion and background clutter Our approach is based on comparing features extracted from views such as intensity edges rather than directly comparing the views themselves The un derlying comparison measure that we use is the Hausdor fraction as opposed to the sum of squared di erences SSD which is employed by most eigenspace matching techniques The Hausdor fraction is quite insensitive to small varia tions in feature location as well as to the presence of clutter or partial occlusion In this paper we de ne an eigenspace approximation to the Hausdor fraction and present some simple recognition experiments which contrast our approach with prior work on eigenspace image matching We also show how to e ciently incorporate our technique into an image search engine enabling instances from a set of model views to be identi ed at any location translation in a larger