<p>Image ordinal classification (IOC) assigns a discrete yet ordinal scalar label to an input image, such as age estimation. Learning similarity metric, as a fine separation tool, is a prevailing solution to this problem. However, it is non-trivial to effectively incorporate the ordinal information of inputs into the classification task, and it is overlooked recently. In this work, we adopt the concept of Mixing techniques, including Mosaic, CutMix and Mixup, which are powerful data augmentation strategies that blend two or more samples and their corresponding labels in a proportional manner during the training process, to bolster the task of ordinal classification. In particular, we delve into the mixing policy from the perspective of metric learning instead of data augmentation, which naturally and simply implements classification learning and difference regression learning, and propose a novel Competitive Learning (CL) prototype for IOC. Firstly, we embed the ordinal measurements of two input images into the training procedure by a virtual combination of image and label, to enhance the discriminative power of local features. We also conduct an in-depth analysis on the impact and advantages associated with implementing mixing policy. Secondly, the difference between two ordinal values is utilized to help measure the subtle difference in IOC, which is integrated with mixing policy properly. Finally, for further improvement, dual augmentation and random augmentation are proposed to enlarge and diversify the disparity of two objects, so as to exploit more robust feature. Comprehensive experiments on two ordinal classification tasks, i.e., age estimation and car date estimation, demonstrate the effectiveness and robustness of our novel approach, which outperforms previous methods with a relatively large margin.</p>