To improve retrieval accuracy and efficiency of the image shape information, this paper proposes an image retrieval method based on the omnidirectional shape feature code. This method segments the shape omnidirectially, measures the shape complexity of every direction and part, constructs the omnidirectional feature codes of the shape, and calculates the similarity of shapes. The actual structure shape data set and MPEG-7 CE-1 Part B shape data set are used to test the retrieval performance of this method, and this method is compared with other description methods of shape similarity. Experimental results show that this method has higher retrieval accuracy and efficiency. The image retrieval method based on the omnidirectional shape feature code can describe exactly global features and partial features of the shape. This method is invariant to translation, rotation, scaling, and strong ability to describe and recognize the shape.