Eosinophilic gastroenteritis (EG) is a gastrointestinal disease characterized by an increase in peripheral blood eosinophil (EOS). The main diagnosis of EG is based on whether the number of eosinophils in the pathological section of a digestive tract mucosa specimen exceeds the standard. In this study, a computer image analysis algorithm was used to identify and count eosinophils in pathological section images, with the aim to assist pathologists to manually calculate the number of EOS and reduce the workload and improve the work efficiency of doctors. A robust watershed algorithm was used as the core algorithm for identifying EOS, and the over-segmentation problem in the traditional watershed algorithm was solved using an improved distance transform algorithm and foreground and background markers. The accuracy of the watershed algorithm for recognition and counting was improved. The improved watershed algorithm was used to identify and count EOS, and its results were compared with a pathologist's standard. The average accuracy of the algorithm is 95.0%. Compared with the traditional watershed algorithm, the relative standard deviation of the improved algorithm improved from 5.8% to 2.2%, the over-segmentation rate reduced from 13.4% to 3.7%, and the running time of the algorithm reduced from 40 s to about 27 s.