Algorithm 3. Working principles of structural risk minimization.

input; image data—{(a1, b1)…, (at, bt)}, f(a)-function, E(f) = expected risk function, generate expected risk function E(f) = $\int Q\left(b,f\left(a\right)\right)dq\left(a,b\right)$ for unknown problem           find empirical risk function           E(f) ← Eemp(f)           Eemp(f) = 1/t${{\displaystyle \sum }}_{k=1}^{t}Q\left(b_k,f\left(a_k\right)\right)$           for nonnegative set                      if (0 ≤ Q (b, f(a)) ≤ Y)                                P ← 1− η                                else if (Q (b, f(a)) ≤ Y)                                 E(f) $\le$ Eemp(f) + $\frac{Y\zeta }{2}(\sqrt{1+\frac{4E_{emp }\left(f\right)}{Y\zeta }})$                       end if                       $\zeta$ ← 2(ln n-ln η)/L                      if L/d large                                Eemp(f) ← small                                E(f) ← small                       end if           end for end for return, minimization of structural risk