Optimization of Image Compression Using K-Means Clustering for Digital Heritage Archives

Preserving digital cultural assets requires efficient compression to minimize storage and bandwidth costs. However, existing studies rarely evaluate K-Means Clustering on structurally complex objects such as the Prambanan Temple, leaving a research gap in assessing its performance against standard codecs. This study introduces a novel optimized K-Means pipeline with adaptive cluster selection and improved centroid initialization for compressing high-detail temple imagery. The method groups pixels based on color proximity, reducing redundancy while preserving key structural patterns. Experiments show that K-Means achieves PSNR 28.08–30.65 dB and SSIM 0.86–0.92, outperforming baseline JPEG at similar file sizes PSNR 26–28 dB, SSIM 0.80–0.87. This quantitative comparison demonstrates the model’s superior perceptual retention in textured stone regions. The methodological contribution lies in combining spatial–chromatic feature weighting with iterative centroid refinement, which increases cluster stability and reduces quantization artifacts. Findings confirm K-Means as a viable alternative for controlled-distortion compression. In conclusion, the proposed approach provides practical engineering implications, enabling reduced storage footprints, predictable reconstruction quality, and integration into hybrid compression pipelines for large-scale digital imaging systems.