We present a novel method for image encryption by Modified Gerchberg–Saxton Iterative Algorithm (MGSIA) using deterministic masks (DMKs) and singular value decomposition (SVD) in fractional Hartley transform (FrHT) domain. This proposal delivers triple-level encryption (L1, L2 and L3) and all these three levels can be iterated number of times. First and third level of encryption routines MGSIA, DMK and FrHT, whereas mid-level encryption uses SVD. Since L1 delivers single cipher image and L3 yields double cipher images, this technique is also known as multiple image encryption system. Each MGSIA generates private keys and these keys are engendered in such a way that decryption is to be expected with traditional DRPE system. Because of the nature of these two asymmetric keys, the proposed system is non-linear and by the way, it boosts security. In the place of traditional random phase masks, the proposed cryptosystem uses statistically independent deterministic masks in L1 and L3. DMKs are nothing but the class of structured phase masks. DMKs are generated and used as the phase for MGSIA in L1 and L3. These DMKs are formed by the deviation from conventional rectangular function and limited range values which delivers key components with reduced size, better performance and lower complexity. Deployment of SVD on L2 decomposes the cipher image1 into three components and out of these three components, due to the nature of diagonality, S components plays a major role in proposed work. Numerical simulations have been achieved to prove the strength and feasibility of the proposed scheme.