Phase retrieval methods used in computer generated holograms (CGH) such as Gerchberg-Saxton (GS) and gradient descent give results which are prone to noise and other defects. This work builds up on the idea of time-averaging multiple hologram frames, first introduced in methods like One-Step Phase-Retrieval (OSPR) and Adaptive One-Step Phase-Retrieval (AD-OSPR). The proposed technique called Multi-Frame Holograms Batched Optimization (MFHBO) uses the L-BFGS optimization algorithm to simultaneously generate a batch of binary phase holograms which result in an average reconstructed image of improved fidelity and fast algorithmic convergence, both in the Fraunhoffer and the Fresnel regimes. The results are compared to OSPR and AD-OSPR in simulation and experimentally, proving the superiority of the proposed approach. This technique can be easily extended to other spatial modulation methods.