Article Preview
TopThe subsequent section covers a short and concise literature review on the Zernike complex moment utilized as features for the recognition characters. It also provides a concise literature survey on the Genetic Algorithm adopted as a feature selection method.
Ghosh & Ghosh (2005) projected a novel scheme with the fusion-based evolutionary technique by utilizing matrix-based solution methods of GAs. Structural features have been utilized as a descriptor of handwritten characters of English script and recognition rates with Backpropagation Nural Network (BPNN) and Evolutionary Algorithm and Least Squares (EALS-BT) algorithm were found 86.4% and 95.6% respectively. Kimura et al. (2009) projected an approach utilizing GA to extract an appropriate group of features out of the given huge pool of features to enhance classification performance.
There are image processing problems which have been successfully resolved using the Zernike Complex Moments. Zernike Complex Moments are also found very suitable as a fundamental instrument to produce shape information of objects in the images, as Zernike Complex Moments hold rotational invariance property, which is very appropriate for off-line character classification/ character reconstruction as discussed in (Teh & Chin, 1988) and (Tripathy, 2010). A different Computer-aided Diagnosis (CADx) scheme to analyze breast masses was suggested (Tahmasbi et al., 2010, 2011). It was intended to enhance the CADx system’s efficacy and lower down False Negative Rate (FNR) through Zernike Complex Moments as a descriptor of shape and edge characteristics of medical images.
The competence of Zernike Complex Moments (ZM) in the development of the image classiðcation model was analyzed by Oluleye et al. (2014). A neuro-genetic intelligent scheme was designed by utilizing a probabilistic neural network (PNN) based classiðer. Geometric features and ZMs were utilized with a new fitness function to discover the best features for optimum recognition rate and attained 90.05% accuracy. Oluleye et al. (2014) further discussed the use of GA in selecting the appropriate features. Explicitly, a binary GA with a new fitness function was adopted to reduce dimensionality to enrich the efficacy of classifiers.
The usability of Zernike Complex Moments as a feature descriptor to recognize the Marathi script’s handwritten compound characters was discussed by Kale et al. (2014). In this study, Support Vector Machines (SVMs) and k-nearest neighbor (k-NN) were utilized as classification techniques and claimed accuracy of 98.37% and 95.82%, respectively. Shahamat & Pouyan (2015) projected a new procedure to classify subjects into two groups characterizing schizophrenia and control utilizing data of functional magnetic resonance imaging (fMRI). Local Binary Patterns (LBP) technique was adopted in extracting the features. GA was employed for feature selection to reveal a feature set with more discrimination strength. Further, for feature selection, linear discriminant analysis (LDA) was employed to extract the features that maximize the ratio between inter-class and intra-class variability.