Luminance correction in stereo correspondence based structure from motion

This paper addresses disparity estimation from image pairs by error minimization. When taken from a monocular camera pan sequence, those images often suffer from luminance variations, caused by the spatially and temporally separated camera positions. Therefore, here we discuss and analyze an existing correction approach for the block matching algorithm. From the observations made we derive a new approach for illumination correction of local error minimization in disparity space. With its simplicity, this multiple disparity space prescaling not only provides the high speed needed for mobile devices, but also decreases the number of false matches by an a priori restriction of correction values.