Abstract:In this paper, an optimal clustering technique for handwritten Nandinagari character recognition is proposed. We compare two different corner detector mechanisms and compare and contrast various clustering approaches for handwritten Nandinagari characters. In this model, the key interest points on the images which are invariant to Scale, rotation, translation, illumination and occlusion are identified by choosing robust Scale Invariant Feature Transform method(SIFT) and Speeded Up Robust Feature (SURF) transform techniques. We then generate a dissimilarity matrix, which is in turn fed as an input for a set of clustering techniques like K Means, PAM (Partition Around Medoids) and Hierarchical Agglomerative clustering. Various cluster validity measures are used to assess the quality of clustering techniques with an intent to find a technique suitable for these rare characters. On a varied data set of over 1040 Handwritten Nandinagari characters, a careful analysis indicate this combinatorial approach used in a collaborative manner will aid in achieving good recognition accuracy. We find that Hierarchical clustering technique is most suitable for SIFT and SURF features as compared to K Means and PAM techniques.
This paper provides an early attempt to train and retrieve handwritten Nandinagari characters using one of the latest techniques in visual feature detection. The data set consists of over 1600 handwritten Nandinagari characters of different fonts, size, rotation, translation and image formats. In the Learning phase, we subject them to an approach where their recognition is effective by first extracting their key interest points on the images which are invariant to Scale, rotation, translation, illumination and occlusion. The technique used for this phase is Scale Invariant Feature Transform (SIFT). These features are represented in quantized form as visual words in code book generation step. Then the Vector of Locally Aggregated Descriptors (VLAD) is used for encoding each of the Image descriptors in the database. In the recognition phase, for query image, SIFT features are extracted and represented as query vector .Then these features are compared against the visual vocabulary generated by code book to retrieve similar images from the database. The performance is analysed by computing mean average precision .This is a novel scalable approach for recognition of rare handwritten Nandinagari characters with about 98% search accuracy with a good efficiency and relatively low memory usage requirements.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.