An auto-restraint-based method for reducing abnormal points of images of sensor under-sampling in CT projection domain is proposed. Decide abnormal points in short-penetration distance which is caused by insufficient sensing and projection data areas has much more different values of noise

so reasonable thresholds is automatically selected according to judged different value point

and carry out filtering

thereby reduce set of empirical value and increase level of auto-restraint noise

moreover statistical distribution is more close to non-stationary Gaussian distribution. Simulation show that the method has a better processing effect for reducing under sampling noise caused by low-dose CT

especially problems of overlarge and too small filtering caused by aging equipment and experience settings.