*Corresponding author. fax: #91-11-6966606. E-mail address: santanu@ee.iitd.ernet.in (S. Chaudhury) Pattern Recognition 32 (1999) 1737 } 1749 Recognition of partially occluded objects using neural network based indexing Navin Rajpal, Santanu Chaudhury*, Subhashis Banerjee Department of Computer Science and Engineering, IIT Delhi, New Delhi, India Department of Electrical Engineering, IIT Delhi, Hauz Khas, New Delhi, 110016, India Received 23 January 1997; received in revised form 9 November 1998; accepted 9 November 1998 Abstract In this paper, a new neural network based indexing scheme has been proposed for recognition of planar shapes. Local contour segment-based-invariants have been used for indexing. Object contours have been obtained using a new algorithm which combines advantages of region growing and edge detection. Neighbourhood constraints have been applied on the results of indexing for combining hypotheses generated through the indexing scheme. Composite hypotheses have been veri"ed using a distance transform based algorithm. Experimental results, on real images of varying complexity of a reasonably large database of objects have established the robustness of the method.  1999 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved. Keywords: Object recognition; Invariant indexing; Neural networks; Hypothesize-and-test; Contour segments 1. Introduction Object recognition is "nding applications in the newer areas of content-based image retrieval, video indexing, etc., in addition to classical application domains like industrial inspection and robotic manipulation. These applications necessitate development of robust recogni- tion techniques which can work with uncalibrated views of the scene and can handle reasonable degree of occlu- sion of the objects. Motivated by these considerations, in this paper, we present an object recognition scheme based on local invariants. Main features of our recogni- tion scheme are extraction of local invariant features by mapping contour segments between dominant points to a canonical frame, indexing based on these features, gen- eration of composite hypothesis about the objects pres- ent in the scene and their pose using neighbourhood constraints on the results of indexing and subsequent veri"cation using an algorithm based on weighted dis- tance transform [1]. Silhouette is extracted from the image using a new pre-processing technique which com- bines advantages of region growing and edge detection [2,3]. Key contribution of this paper is the use of an indexing mechanism in which the capabilities of neural network have been exploited for computing association between image features and object models in a robust and e$cient fashion. The majority of techniques [1}6] which are capable of recognizing occluded objects of di!erent sizes and ori- entations work well for model base containing small number of objects. For a model base containing large number of objects, invariant-based indexing schemes have been proposed in the literature [7,8]. In all these methods, an index is typically a pattern vector of invariant 0031-3203/99/$20.00  1999 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved. PII: S 0 0 3 1 - 3 2 0 3 ( 9 8 ) 0 0 1 6 4 - 2