Algorithm 1: Discriminative classification of target objects.

Input: frame images $I_{1:t_0}$, states $X_{1:t_0}$, integers $r_{tg},r_{bg}>0$  1.take target patches ${\overrightarrow{p}}_{1:t_0}^{target}$  2.take background patches ${\overrightarrow{p}}_{t,(a_x,b_y)}^{back},t=1,\dots ,t_0$  3.create target appearance descriptors ${\overrightarrow{v}}_{1:t_0}^{tg}\in {\mathbb{R}}^s$  4.create background appearance descriptors ${\overrightarrow{v}}_{1:\kappa }^{bg}\in {\mathbb{R}}^s$  5.find $(D,C)$ by applying structured sparse PCA (6)  6.find optimized classifier $\mathsf{\Phi }$ such that $\mathsf{\Phi }({(DC)}_i)\ge 1$ for $i=1,\dots ,t_0$ and $\mathsf{\Phi }({(DC)}_i)\le -1$ for $i=t_0,\dots ,t_0+\kappa$

Output: target appearance descriptors ${\overrightarrow{v}}_{1:t_0}^{tg}\in {\mathbb{R}}^s$ and classifier $\mathsf{\Phi }$