A CMOS image sensor with motion vector estimator for low-power image compression

This paper presents a novel method of motion vector estimation (MVE) based on the integration of a CMOS image sensor and a motion vector estimation on a chip. The CMOS image sensor has two sensing modes; normal destructive video-rate (30 frame/s) image sensing mode and nondestructive high-speed interpolated image sensing mode. The nondestructive image-sensing is essential to obtain high-speed interpolated images while maintaining high image quality video-rate image. A novel iterative block matching algorithm allows us to estimate the video-rate (30 frame/s) motion vectors accurately from the motion vectors obtained at high-speed interpolated pictures. The iterative block matching reduces the computational complexity by a factor of more than one-tenth compared with the conventional full search block matching algorithm, allowing totally low-power design of video encoders