Digital subtraction radiography is a powerful technique for the detection of changes in serial radiographs. Among the others, contrast correction is a basic step for comparing the radiographs. Ruttimann's algorithm is widely used for contrast correction. In this study we propose a technique which is based on smoothing the empirical distribution of the reference image to improve Ruttimann's algorithm. Cardinal splines were used for smoothing the empirical distribution. Results based on clinical and simulated data showed that the proposed technique has outperformed the Ruttimann's algorithm. Relationship between the color depth and contrast differences was also investigated in terms of peak to signal ratio metric.