The image encryption technology is an important means to ensure network digital data transmission security. To improve the security and encryption computational efficiency, we proposed a novel bit-level image encryption algorithm that integrates hyper-chaotic systems with DNA encoding operations. First, we generated high-quality pseudo-random sequences through K-means clustering, which were applied to hyper-chaotic sequences; second, we implemented multi-level scrambling at both bit-plane and bit-level; finally, we employed dynamic DNA encoding rules combined with bidirectional diffusion. Experimental results demonstrated that the proposed algorithm achieves a large key space (approximately $2^{256}$), high information entropy (close to 7.9899), and strong resistance to differential attacks (NPCR = 99.61%, UACI ≈ 33.46%). Our results also showed that the histograms of the encrypted images are even, and the correlation coefficients approach 0. Moreover, the proposed algorithm can effectively resist the chosen plaintext attack, brute force attack, clipping attack, salt and pepper noise attack, etc.