Multimodal Recognition Framework: An Accurate and Powerful Nandinagari Handwritten Character Recognition Model
Prathima Guruprasad, Jharna Majumdar · Procedia Computer Science · 2016
Effective and efficient recognition of ancient Nandinagari Handwritten manuscripts depend on the identification of the key interest points on the images which are invariant to Scale, rotation, translation and illumination. Good literature for finding specific types of key interest points using single modality is available for scripts. However, finding the best points and its descriptors is challenging and could vary depending on the type of handwritten character. Its choice is also dependent on the tradeoff between precision of recognition and speed of recognition and the volume of handwritten characters in question. Thus for a recognition system to be effective and efficient, we need to adopt a multimodal approach in which the different modalities are used in a collaborative manner. In this paper, instead of separately treating the different recognition models and their algorithms, we focus on applying them at a phase where their recognition is optimal. A varied set of Handwritten Nandinagari characters are selected with different image formats, scale, rotation, translation and illumination. Various models of feature extraction techniques are applied of which SURF is chosen to be very performance efficient with good accuracy. A dissimilarity ratio matrix is computed out of the maximum number of matched features of query to test image and vice versa to denote the magnitude of dissimilarity between the images in the data test updated with the new query image. The result is processed through agglomerative clustering to achieve powerful and accurate clusters belonging to designated set of images with 99% percent recognition.