Person re-identification (re-ID) includes two vital parts: feature representation and metric learning. However, person re-ID is affected by cross-camera factors such as illumination, pose and viewpoint variations. In this paper, we propose a novel feature called Global Maximal Occurrence (GOMO) based on Local Maximal Occurrence (LOMO), and an enhanced metric learning method. Our proposed feature not only maximizes horizontal occurrence of local features but also combines the maximum occurrence of vertical and horizontal features. To handle vertical feature misalignment between the cross-cameras, we maximize the maximum occurrence of the vertical direction and then cascade features in two directions as a fusion feature. Besides, histogram equalization is applied to enhance images for overcoming cross-camera variations. To reduce the errors between camera views caused by the outliers in final distance matrix of traditional metric learning, we propose a re-ranking method using the max-min criteria that linearly map all values in final distance matrix to a specified subspace to enhance the Cross-view Quadratic Discriminant Analysis (XQDA) method. We conducted experiments on two challenging person re-ID datasets, VIPeR and CUHK Campus, and the experimental results demonstrate the effectiveness of our proposed feature and the superiority of our enhanced metric learning method over the state-of-the-art methods.