Non-parametric estimation of camera function

In this paper a non-parametric, direct estimation of camera response function is presented and evaluated in real and simulated data, using multiple images of an arbitrary scene acquired in different exposure times. Initially, it is proved that the subtraction of the logarithm of the inverse response function of any pair of pixels is invariant to the integration time and equal to the logarithm of the ratio of the camera responses. Based on this property, the proposed method estimates the response function in a non-iterative manner, solving an appropriately set linear system of equations. The direct solution, evaluated under a representative set of experiments, shows that even in the case of unknown exposure times, the mean absolute error of the camera´s estimated response function is lower than 0.020 in all simulation experiments and 0.018, 0.021, and 0.023 for the RGB channels of the Olympus 5050Zoom digital camera