A Robust Texture Preprocessing for AAM
Mingcai Zhou, Yangsheng Wang, Xuetao Feng, Xiaoyan Wang · 2008
The Active Appearance Model (AAM) is a powerful generative method for modeling and registering deformable objects. However, model based on raw intensity tends to be sensitive to changes in conditions such as lighting. Thus model built on one data set may not perform well on another data set taken under different conditions. In this paper, a new low-cost preprocessing method for the local gradient is proposed. The proposed preprocessing neglects the edge direction but accentuates the strength of likely edges. In particular, the algorithm firstly computes the square of each component of the local gradient and then applies a non-linear normalization to them respectively. Experimental results demonstrate that our preprocessing method can achieve higher accuracy and better convergence rates.