Abstract: When ear posture varies greatly, the ear recognition rate decreases significantly. In this paper, a general recognition method based on local tangent space alignment is proposed to extract two-dimensional pose-invariant ear features. Unlike LTSA constructing local tangent space point by point, this method presented a novel strategy to construct several overlapped local maximum linear patches to approximate the nonlinear multi-posed ear manifold. The number of the patches was far less than the total amount of dataset. Then these linear patches were aligned to obtain the corresponding lower dimensional manifold. The experimental results on USTB ear database reveal that the proposed method can achieve better recognition performance and has less computation. It can be an effective multi-posed image recognition method.