Embodiments of the present disclosure disclose a method and an apparatus for processing an image. A specific embodiment of the method includes: acquiring a target image including a polygon image; inputting the target image into a pre-trained convolutional neural network to obtain a characteristic vector of the target image, the convolutional neural network being used to represent a correspondence relationship between an image and a characteristic vector, and the characteristic vector being a vector including a category feature, a position feature, and a keypoint feature of the polygon image; and recognizing an image area of the polygon image based on the category feature, the position feature, and the keypoint feature. This embodiment improves the accuracy of polygon image recognition.