A horse face available weighted-constructed facial image recognition approach in low-dimensional space

A facial image recognition method combining weighted function and maximum-minimum-median average MSD-based (A3MMSD) is developed. The main weakness of existing A3MMSD model is that the class-vector in the expressions of within-class scatter matrix (Sw) and between-class scatter matrix (Sb) is estimated by maximum-minimum-median average (A3M). For a few of given samples with non-ideal conditions, such as variations of illumination and pose, the assessment result is very weak by using the class sample A3M-vector. Therefore, the recognition performance of A3MMSD criterion will decline significantly. In order to solve this problem, within-class weighted-A3M vector is used to construct new within-class scatter matrix and new between-class scatter matrix, that is the original within-class A3M-vector will be replaced by a new created vector. The proposed method integrates the merits of weighted function and A3MMSD method. The weighted function based on Euclidean distance adds class differences in the low-dimensional space to obtain the optimal projection matrix. Experimental results show that an improvement of the A3MMSD model is possible with the proposed technique in ORL and horse facial image dataset recognition problems.