Person Re-identification (ReID) aims to retrieve a target pedestrian from an image gallery captured by cameras in varied scenarios. It is crucial for ReID to extract extensive discriminative feature representations from images for achieving desirable performance. The majority of current methods focus on mining data that can identify a pedestrian from a single image by investigating different dimensions of the image. However, a single image is sometimes insufficient to precisely characterize all the necessary features for identifying a pedestrian especially when the data quality is not guaranteed. Since a pedestrian tends to be caught in numerous images, information missed in a single image is expected to be supplemented from other images. Therefore, we consider extracting more robust feature representations benefiting from relationships between multiple pedestrian images and propose a new method DTMIReID. Firstly, we suggest a Dual Branch Attention Module (DBAM) based on Transformer to extract global and local features from single images. Then we combine the extracted features of multiple images together and input them into our proposed Deformable Transformer Module (DTM) to simultaneously fuse the global and local features from these multiple images by a Sample-Points-Based Attention (SPBA) mechanism. To the best of our knowledge, our method is the first ReID model that uses the Deformable Transformer to establish relationships between multiple features. Experimental results on four large ReID datasets show that the new method outperforms state-of-the-art published works by a large margin. DTMIReID is available at https://github.com/Titaniumyh/DTMIReID.git .