The problem of offline handwritten Chinese character recognition has been extensively studied by many researchers and very high recognition rates have been reported. In this paper, we propose to further boost the recognition rate by incorporating a distortion model that artificially generates a huge number of virtual training samples from existing ones. We achieve a record high recognition rate of 99.46% on the ETL-9B database. Traditionally, when the dimension of the feature vector is high and the number of training samples is not sufficient, the remedies are to (i) regularize the class covariance matrices in the discriminant functions, (ii) employ Fisher's dimension reduction technique to reduce the feature dimension, and (iii) generate a huge number of virtual training samples from existing ones. The second contribution of this paper is the investigation of the relative effectiveness of these three methods for boosting the recognition rate.

The topics of fingerprint classification, indexing, and retrieval have been studied extensively in the past decades. One problem faced by researchers is that in all publicly available fingerprint databases, only a few fingerprint samples from each individual are available for training and testing, making it inappropriate to use sophisticated statistical methods for recognition. Hence most of the previous works resorted to simple-nearest neighbor (-NN) classification. However, the-NN classifier has the drawbacks of being comparatively slow and less accurate. In this paper, we tackle this problem by first artificially expanding the set of training samples using our previously proposed spatial modeling technique. With the expanded training set, we are then able to employ a more sophisticated classifier such as the Bayes classifier for recognition. We apply the proposed method to the problem of one-to-fingerprint identification and retrieval. The accuracy and speed are evaluated using the benchmarking FVC 2000, FVC 2002, and NIST-4 databases, and satisfactory retrieval performance is achieved.

No abstract

Abstract. Middleware support is a major topic in pervasive computing. Existing studies mainly address the issues in the organization of and the collaboration amongst devices and services, but pay little attention to the design support of context-aware pervasive applications. Most of these applications are required to be adaptable to dynamic environments and self-managed. However, most context-aware pervasive applications nowadays have to carry out tedious tasks of gathering, classifying and processing messy context information due to lack of the necessary middleware support. To address this problem, we propose a novel approach based on ontology technology, and apply it in our Cabot project. Our approach defines a context ontology catered for the pervasive computing environment. The ontology acts as the context information agreement amongst all computing components to support applications with flexible context gathering and classifying capabilities. This allows a domain ontology database to be constructed for storing the semantics relationship of concepts used in the pervasive computing environment. The ontology database supports applications with rich context processing capabilities. With the aid of ontology technology, Cabot further helps alleviate the impact of the naming problem, and support advanced user space switching. A case study is given to show how Cabot assists developers in designing context-aware pervasive applications.