Research Areas
  Image Processing >> CIS Video Stabilization  |  Plasma Display Panel  |  Segmentation

CIS Distortion Analysis and its Application to Digital Image Stabilization

Abstract
 In CMOS image sensors (CISs), CIS image distortion is due to the rolling shutter and it can degrade the visual quality of a CIS image when a CIS camera moves rapidly. In this work, we propose the affine motion based CIS distortion correction method combined with the digital image stabilization method. To remove CIS distortions due to the affine global image motion, CIS distortion model is proposed to explain the effect of the rolling shutter mechanism. To improve CIS video's visuals, we propose CIS digital image stabilization to remove the jittering motions in the new image sequence obtained by our correction method

The shuttering mechanism: CCD (Charge-coupled device) vs. CIS (CMOS image sensor)

Fig. 1 The illustration of an image acquisition process in a CCD and a CMOS image sensor

 Since CISs do not have a global shutter like a CCD, the CIS reads out the exposed row one by one to make an image while its other scan lines except for the current read-out row are still being exposed. This is called a rolling shutter.

CIS distortions analysis and its experiments
 We test indoor sequences captured by a 330K pixel CIS USB PC webcam. The video has two parts: the left is the original frame and the other is the undistorted and stabilized video.


Fig. 2 The translational distortion: (top) original CIS frames (bottom) their undistorted frames

Fig. 3 The rotational distortion: (top) original CIS frames (bottom) their undistorted frames

CIS digital image stabilization
 Our CIS digital image stabilization algorithm consists of three parts: a global motion estimation, a CIS distortion correction, and an undistorted CIS stabilization. Since the viewers usually feel uneasy about video¡¯s visuals due to rapid translational motions of a camera, we newly estimate the translational motion between new undistorted CIS frames.

 We test outdoor sequences, video1 and video2 which are captured by a 3.2M pixel CIS handy camcorder. The video has two parts: the left is an original frame and the other is our undistorted and stabilized frame.


Fig. 4 an outdoor test: (top) original frames (bottom) CIS stabilized video result





References
 [1] Won-ho Cho and Ki-sang Hong, ¡°Affine motion based CMOS distortion analysis and CMOS digital image stabilization,¡± will be published in IEEE Trans. on Consumer Electronics, 2007.
 [2] Won-ho Cho, Dae-woong Kim, and Ki-sang Hong, ¡°CMOS digital image stabilization,¡± will be published in IEEE Trans. on Consumer Electronics, 2007.