F. Roli and S. Vitulano (Eds.): ICIAP 2005, LNCS 3617, pp. 37 – 49, 2005. © Springer-Verlag Berlin Heidelberg 2005 Interactive, Mobile, Distributed Pattern Recognition George Nagy DocLab, Rensselaer Polytechnic Institute, Troy, NY USA 12180 nagy@ecse.rpi.edu Abstract. As the accuracy of conventional classifiers, based only on a static partitioning of feature space, appears to be approaching a limit, it may be useful to consider alternative approaches. Interactive classification is often more accurate then algorithmic classification, and requires less time than the unaided human. It is more suitable for the recognition of natural patterns in a narrow domain like trees, weeds or faces than for symbolic patterns like letters and phonemes. On the other hand, symbolic patterns lend themselves better to using context and style to recognize entire fields instead of individual patterns. Algorithmic learning and adaptation is facilitated by accurate statistics gleaned from large samples in the case of symbolic patterns, and by skilled human judgment in the case of natural patterns. Recent technological advances like pocket computers, camera phones and wireless networks will have greater influence on mobile, distributed, interactive recognition of natural patterns than on conventional high-volume applications like mail sorting , check reading or forms processing. 1 Introduction I am grateful for this wonderful opportunity to proselytize for some heretical notions. First, I will suggest classifying pattern recognition applications into types A, B, AB, and O, according to the pattern recognition methodology that suits each best. Type A consists of symbolic patterns, the glyphs and sounds used for encoding messages. Type B includes natural objects like flowers and faces that are not used primarily for communication. I will try to substantiate the claim that interactive computer vision, where the tasks leading to object recognition are assigned according to the relative competence of human and machine, is particularly appropriate for Type B applications. On the other hand, context and style based classification seems better suited to Type A applications. Learning and adaptation benefit every type. Section 2 outlines the considerations that led to the proposed taxonomy of recognition problems. Section 3 summarizes our recent results on interactive classification of flowers and faces. Section 4 and 5 present the corollaries of interactive classification: mobile and networked recognition. In Section 6 we recapture the notion of style, show that it can lead to more accurate classification of multi-source patterns when the test samples are partitioned by source, and contrast it to the better established methods based on language context. In the last section I list some areas where rapid progress may be possible. This is not a survey: however, the cited publications contain extensive references to invaluable prior work by others.