Mixture of trees with three layers

The Tree Structured Model (TSM) is proven effective for solving face detection, pose estimation and landmark localization in an unified model; however, the drawback of its processing time makes it unfavorable in practical applications, especially when dealing with cases of multiple faces. We propose the Mixture of Trees with Three Layers (3L-MoT) to improve the run-time speed. The 3L-MoT is composed of three component TSMs, the coarse TSM (c-TSM), median TSM(m-TSM) and the refined TSM (r-TSM), and a Bilateral Support Vector Regressor (BSVR). The c-TSM is built on the low-resolution octaves of samples so that it provides coarse but fast face detection. The m-TSM remove the false positive generate from c-TSM to speed up the process in r-TSM. The r-TSM is built on the high resolution octaves so that it can locate the landmarks on the face candidates given by the m-TSM and improve precision. The r-TSM based landmarks are used in the forward BSVR as references to locate the dense set of landmarks, which are then used in the backward BSVR to relocate the landmarks with large localization errors. The forward and backward regression goes on iteratively until convergence. In spite of the negative correlation between run-time speed and performance, the performance of the 3L-MoT is validated on Multi-PIE benchmark databases to be similar with TSM[10] while enhanced in processing time.