Biometric recognition is one of the effective authentication techniques which is utilized in various applications for making the individual identification process. During the verification and authentication process different biometric features such as signature, ear, iris, face, palm, finger knuckle details are used to perform this process. Due to the easy acceptance of the palm surface, fine textures and stable features characteristics are helps to choose the finger knuckle feature for biometric process in this work. First the finger biometric features are collected from PolyU finger knuckle database. After that, the noise present in the images are eliminated using weighted median filter and the knuckle region is located with the help of the variational approach. After that key point descriptors are extracted using sparse autoencoder approach. Finally, the specific features are trained using compositional networks and features matching is performed by Chebyshev distance. The matching process authenticate the user whether they are authorized person or not. At last efficiency of the system is evaluated using MATLAB based experimental results such as false acceptance rate, equal error rate and false rejection rate.