To address the issues of sluggish processing speed and substandard quality of the halftone effect in image halftoning, an efficient halftone algorithm was proposed, which relies on a lightweight residual convolutional neural network(CNN). A noise compensation block was introduced to provide the model with jitter dependencies, to address the issue of flatness degradation in the original CNN. The inclusion of blue noise loss in the loss function was implemented to further optimize the model’s performance. The low-frequency components were suppressed and the anisotropy of the high-frequency region was optimized when halftoning a constant-value grayscale image. Experimental results showed that, compared with the available deep halftone methods, the proposed algorithm exhibited a significant reduction of 96.77% in the number of parameters, the structural similarity (SSIM) improved by 8.17% and the peak signal-to-noise ratio (PSNR) improved by 0.133 3 dB in the VOC test set, the halftone images had blue noise characteristics, and the processing speed was increased by 57.28%.