With the rapid expansion of digital communication, protecting sensitive visual data, including personal identifiers, medical images, and confidential corporate content, has become increasingly critical, especially as advancements in quantum computing threaten the security of classical cryptographic techniques. To address this challenge, we introduce a quantum-resistant image encryption framework that combines adaptive signal transforms with generative learning for enhanced robustness and security. The core of the proposed method is the Improved Rajan Transform (IRT), a signal-invariant transformation designed to manipulate pixel-level features while preserving directional invariance. The IRT is augmented with a Cumulative Point Index (CPI) and chaotic permutation through the Arnold Cat Map, providing significant spatial scrambling. For diffusion, an XOR-based mechanism driven by transformation matrices generated via the Henon map increases entropy and key sensitivity. To further strengthen defenses against quantum and statistical attacks, a lightweight generative adversarial network (GAN) dynamically generates pseudo-random key matrices using image content and entropy feedback loops, enabling intelligent adaptation to varying image complexities. This hybrid design seamlessly integrates chaos theory, advanced signal processing, and machine learning, offering a resilient encryption system capable of resisting brute-force decryption even in post-quantum scenarios. Experimental evaluations demonstrate superior performance, with metrics such as Correlation Coefficient (CC), Number of Pixels Change Rate (NPCR), and Unified Average Changing Intensity (UACI) confirming strong resistance to statistical, differential, and cryptanalytic attacks. By merging chaos-based transformations with AI-driven key generation, the proposed approach delivers a scalable, adaptable, and forward-looking encryption solution, setting a foundation for secure and future-proof image communication.