In the presentera, human authentication is tightly coupled with biometric advancements. Authentication deals with metrics of both physical and behavioral traits of an individual to ensure strong acknowledgment plans by either accommodating or deciding the intentions of an individual based on their assistances. The key motivation behind such plans is to guarantee that human authentication is done with high degree of certainty and confirmation by adhering to factors like accuracy, platform adaptability, speed, usage and versatility whilst balancing long-term issues of database size and security aspects. Eye biometric has demonstrated itself to be exceedingly flexible and suited for extensive populace and security applications. Security system accomplishes the value for two basic functions: Verification and Identification of users. This paper concentrates on representation and reconstruction of biometric features at minimum cost with deep learners. In this paper, we propose a bimodal recognition system that fuses deep feature representations of iris and retina extracted through Autoencoder (AE) and Gray Level Co-occurrence Matrix (GLCM) respectively at the feature level. The fused deep feature map can be reconstructed from generic features through unsupervised leaner, Autoencoder when stored templates are vulnerable to spoofing. Minimum Cost Matcher (MCM) is employed to enhance the accuracy and efficiency of the bimodal recognition system. The experimental results demonstrated that proposed system outperforms over the unimodal recognition systems of iris and retina.