Cellular neural networks for motion estimation

Parallelization of the optical flow computation in sequences from moving cameras

“…In some cases, SIMD processor arrays with specific chips, either existing [36] or designed ad hoc for the computation of optical flow [13,17,[37][38][39], have been used. General-purpose MIMD as the connection machine [40,41], networks of transputers [42], or cellular neural networks [43,44] were also used in the past.…”

Video Compression by Neural Networks

“…CNNs perform an optimization process based on their capacity to evolve toward a global minimum state. Figure 5 shows the cell architecture of the network described in [36]. Cells are located in a N M × array; the generic cell C ij has a state x ij , a constant external input u ij and an output y ij and has r neighbor cells.…”

“…In [36] motion estimation is based on maximization of the a-posteriori probability (MAP) of the scene random field given the random motion field realization. It is possible to find similarity between MAP and CNN energy function.…”

“…The architecture of the neural network performing the vector optimization is shown in figure 4. Other approaches to motion estimation include cellular neural networks (CNNs) [36]- [39] [76]. These architectures can parallelize the computational flow required by both motion estimation and compensation, yielding faster and scalable computations.…”

“…Cells are located in a N M × array; the generic cell C ij has a state x ij , a constant external input u ij and an output y ij and has r neighbor cells. [36] It is a graphical representation of the following difference equation:…”