Photographically Guided Alignment for HDR Images

This paper presents an automatic image alignment algorithm that alleviates the need to keep the camera still during the capture of a bracketed sequence to obtain an HDR image. Our algorithm assumes that the misalignment between the two consecutive exposures is translational. Using a photographically guided random search, our algorithm first finds properly exposed high contrast regions. The shift amount is then found by analyzing and matching the pixel correlations inside these regions.