To evaluate the residual load capacity of an existing bridge from point cloud data, it is efficient to create an FE model. In the FE model of steel structures, a fiber-based model is reasonable. In the case of including corroded area, it is desirable to convert them to a shell model because local buckling may be predominant. However, it is not easy to find the corroded area from discrete point cloud data. Therefore, using photogrammetry, it is possible to detect the corroded area with image processing technology. The authors have developed the method to generate a fiber-based model from point cloud data of photogrammetry and confirmed that the model can be used for residual load capacity analysis. In this research, the method is developed to generate an efficient FE model by combining photogrammetry and image processing of a steel specimen with pseudo-corrosion. It generates an FE model consisting of a medium-precision fiber-based model in the intact area and a high-precision shell model in the corroded area. Finally, the specimen is subjected to a bending experiment, and the behavior is compared with the analytical results to verify the validity of the method.