Edge information of a license plate image plays an important role in segmenting and recognizing characters in a license plate. By using the bimodal characteristic of a license plate, we propose an edge-preserving compression method of a license plate image. Our method can be applied to the license plate recognition system where a camera only detects a license plate from an input image, and the server segments and recognizes the characters in the transmitted license plate image. Experimental results show that our method preserves edge information more than the conventional 2D image encoder.