Low-Complexity Colour Image Compression for Resource-Constrained Devices Using 4:2:0 Chroma Subsampling and a DPCM-Conditional RLE-Huffman Cascade

Colour images captured by drones, IoT cameras and embedded sensors must often be stored or transmitted by devices that cannot host transform-based or learned codecs. Lightweight run-length-based schemes suit such devices, but they compress R, G and B independently, ignore the lower sensitivity of human vision to chrominance, and perform well only on images with long runs of identical pixels. This paper presents a low-complexity lossy compression pipeline in which each stage targets one class of redundancy: BT.601 YCbCr conversion separates luminance from chrominance, 4:2:0 subsampling removes perceptual chrominance redundancy, and a per-channel cascade of first-order DPCM, conditional run-length encoding and canonical Huffman coding removes inter-pixel, run-length and symbol-frequency redundancy. The entropy-coding cascade is bit-exact, so all distortion is confined to chroma subsampling and colour-transform rounding. The bitstream is fully specified. On a corpus of 113 images in 5 disjoint content categories, the pipeline reduces images to a mean of 23.27% of their raw 24-bit size (5.58 bits per pixel) at a mean PSNR of 45.0 dB, SSIM of 0.9999 and CIEDE2000 colour difference of 0.61. An ablation study shows that Huffman coding contributes the largest share of the gain in the lossless domain and that the conditional RLE rule prevents expansion on run-poor content. In a same-image comparison on the public Kodak dataset with PNG, JPEG-LS, QOI, WebP, JPEG XL, JPEG and JPEG 2000, the pipeline produces the smallest file of any low-complexity codec tested, and it requires no dictionary or trained model, but only sparse Huffman codebook. Compressed size correlates strongly with the mean absolute DPCM residual (r = 0.81) and with luminance entropy (r = 0.64), making performance predictable from image content.