A low-power video motion estimation array processor

For remote-sensing and portable video encoding applications, reducing power consumption is important due to the limited power budget. In systems such as MPEG, motion estimation (ME) accounts for most of the compression and computation. In ME processing, each image frame is broken into macro-blocks (MB) for searching in the previous or next frame using mean absolute difference (MAD) as the metric function. When a good match is found, the motion vector and error block are sent. Smart searches have fewer MAD operations than brute-force search, saving power with some loss in compression ratio. The brute-force and a variation of 2Log algorithms are compared in this paper.