Even if the preprocessing is discounted, the SAD generally requires more resources than the CT. As shown in Fig. 5, unlike the CT that reuses previously extracted features, the SADs must be recomputed every time. Also, SAD implementations that achieve similar throughput, such as the one in [36], require an additional window buffer to store previous column SAD computations (bottom of Fig. 5). Thus the resource requirement is greater due to both the greater number of computations as well as the greater need for memory.