Low-Complexity Image Compression Using Semantic-Aware Block Truncation Coding

Abstract

Block Truncation Coding (BTC) is a simple and computationally efficient image compression technique, but its fixed quantization strategy often leads to suboptimal perceptual quality, particularly in textured and edge-dominant regions. To address this limitation, this paper proposes a Semantic-Aware Block Truncation Coding (SA-BTC) scheme that integrates local variance as a semantic feature to adaptively control quantization at the block level. By adjusting compression strength according to texture characteristics, the proposed method preserves structural and semantically important details in high-variance regions while applying stronger compression to smooth areas. Experimental evaluations conducted on standard benchmark images demonstrate that SA-BTC achieves improved perceptual reconstruction quality compared to conventional BTC. Although the proposed method exhibits lower peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) values than classical BTC, this behavior is expected due to its semantic-aware design, which prioritizes visually significant regions such as edges and facial features over background fidelity. Visual inspection confirms that SA-BTC better preserves edge continuity, contrast, and meaningful structures, including tripod edges, clothing details, and facial components, despite mild block artifacts in homogeneous regions. These results indicate that conventional pixel-based metrics may underestimate the perceptual and semantic quality achieved by SA-BTC. Overall, the proposed approach enhances rate distortion performance from a perceptual perspective without increasing computational complexity, making it well suited for low resource image compression applications where semantic and visual fidelity are more critical than strict pixel-wise accuracy.