Neural Architecture Using Energy Minimization

The output reflectances are computed by minimizing an energy function that is computed using the data of layers B, C, and D. The energy function E consists of three components E, 1-2- and 3. The relative importance of the components is set by adjusting the weights w 1 and u 2. 

The first component is used to compute the reflectances according to the algorithm of Horn 

Additional neural models for color constancy, which were not discussed in depth, were proposed by Courtney et al. (1995). 

---