Despite years of research in using the fingerprint as a biometric entity, there remains a niche to improve the underlying accuracy, specifically on enhancement, feature extraction and classification consideration. Attempted in this paper is a ridge compensation filtering technique for enhancement, followed by minutiae extraction and false minutiae removal and eventually a procedure for template creation. The extracted features are evaluated using metrics like accuracy, error rate and goodness index. These values were found to be better when compared with the results of some state-of-art algorithms, which endorse the effectiveness of the first two stages. Mahalanobis distance (MD) is used to recognise the fingerprints through the generated features that resulted in accuracy above 92%. Finally, convolution neural network (CNN) is used to cross-examine the recognition capacity. Though CNN also able to give a similar accuracy, the proposed algorithm is simple and robust and hence easily embedded in hand-held devices.