Simultaneous Cascaded Regression

摘要: This paper addresses the problem of localizing facial landmarks with deformable face models using cascaded regression strategies. Recently, these methods have become quite popular, standing out as simple and efficient approaches to optimize nonlinear objective functions. In this paper, we target well-known Lucas Kanade (LK) image alignment formulation introduce Simultaneous Cascaded Regression (SCR) technique, which can be considered a extension Forwards Additive / Inverse Composition approaches. contrast previous LK techniques (Newton based optimizations) require recompute Jacobian Hessians matrices at each iteration, our approach learns (offline) sequence descent directions, effectively behaving averaged steepest matrices. Under revised propose part-based generative model (with linear warp function), that accounts underlying shape appearance structure embedded into process itself. Our method is validated on number experiments several datasets (LFPW, LFW, HELEN, 300W), demonstrating noticeable gain in accuracy/fitting performance when compared other solutions.