The Effect of Radiometric Correction on Multicamera Algorithms

We present results confirming the importance of radiometric correction in multicamera applications. Although, we compensate for systematic noise only, we review all noise sources in the video sensor (systematic and random). We use a simple model for radiometric correction of digital images. The correction procedure is tested on the disparity map computation in stereo matching, particularly in a case where stereo usually fails -- almost textureless white surface. Without correcting radiometricly, the matching algorithm matches systematic noise components in the two images. With the correction, after removing the systematic noise, an improvement of 26% to 59% in relative rms of the disparity map is demonstrated (the higher the intensity of the flat field, the better the improvement)