There is a large amount of chemical structure stored in image format which is in the scientific literature and the network. Researching on how to recognize chemical structure from images is contribute to storage management of relevant images. The overall recognition effect of chemical structure has room for improvement. This paper proposes a new recognition for two-dimensional organic chemical structure. Firstly, preprocess the image, and then recognize and remove the isolated chemical symbols and dashed bonds. Then vectorize the remaining images and recognize the accretive chemical symbols and chemical bonds from the vectorization results. Finally, repair the results of misidentification and reorganize the chemical symbols and chemical bonds. The experiments show that the recognition rate in this paper is higher than the existing universal recognition method.