Deep Neural Networks (DNNs) remain vulnerable to backdoor attacks. These attacks are pernicious when attackers inject a trigger into the input data without altering its ground-truth label, known as clean-label backdoor attacks. The effectiveness and stealth of existing clean-label backdoor attacks rely on access to global training data, making them resource-intensive and impractical. This paper introduces a novel clean-label backdoor attack that achieves high attack success rates (ASR) while maintaining exceptional stealth under realistic constraints. Unlike prior approaches focused on spatial differences between clean and poisoned data, a key novelty of the new attack is the suppression of high-frequency components in the frequency domain, which minimizes human-detectable contrasts and enhances trigger imperceptibility. Moreover, we employ an affine combination of static and adaptive triggers, effectively balancing their strengths to maximize ASR. Surrogate models are designed to simulate victim model behavior and auxiliary models are applied to extract the spatial and frequency features of the static triggers, enabling trigger design with minimal knowledge of the victim model. Our attack achieves a higher ASR than state-of-the-art backdoor attacks while maintaining higher benign accuracy across various models and datasets. It also demonstrates strong resistance against the latest defense mechanisms, including STRIP, SentiNet, neural cleanse, fine-pruning, and ANP.