A novel algorithm for segmentation and classification of Malayalam characters through characterization using dominant singular values

The rapid process of automation in recent times involves the development of efficient algorithms in machine intelligence. Man-machine interface in most areas of application involves vision systems to recognize texts and objects. In either case this finally reduces to the machine analysis of images ultimately leading to classification of images either as a whole or in segmented form. Images are represented as matrices in some form or the other. Characterization of matrices for efficient classification is a challenging task. Reduction of dimensionality is important in applications that need to handle large number of such matrices. Image recognition and in particular character recognition is an excellent example where large number of image matrices need to be stored and retrieved often at high speed, at the same time performing computational tasks, resulting in requirements of huge memory and computation time. Characterization in various forms could result in reduction in dimensionality to different extents. This paper proposes the use of dominant singular values as distinct characteristic features of rectangular image matrices suitable for reliable classification. Several analysis and outcomes are reported in this paper. A recognition rate of 97.61% has been achieved for the recognition of printed Malayalam characters.