The total variation (TV) minimization algorithm is an effective CT image reconstruction algorithm that can reconstruct sparse or noisy projection data with high accuracy. However, in some cases, the TV algorithm produces stepped artifacts. The relative TV algorithm outperforms TV algorithm in the field of image denoising. In view of this, the relative TV model is introduced into image reconstruction, a relative TV minimum optimization model is proposed, and the corresponding solution algorithm is designed under the framework of adaptive gradient descent projection to the convex set (ASD-POCS) to further improve reconstruction accuracy. The reconstruction experiments were conducted with Shepp Logan, Forbild, and real CT image simulation models to verify the anti-crime ability of the algorithm and evaluate the sparse reconstruction and anti-noise abilities of the algorithm. The experimental findings reveal that the algorithm outperforms the TV method in terms of anti-crime capability and the ability to reconstruct sparse or noisy projection data with high precision. Compared with the TV algorithm, the algorithm can achieve higher reconstruction accuracy.