The work aims to study discriminative power and retrieval performance of shape feature of the objects in image description and image retrieval. An image retrieval method synthesizing PHOG shape and lifting-based wavelet transform was designed. Firstly, the orientation normalized polar coordinate system of the original image was achieved to extract the image rotational invariant features. Secondly, PHOG shape feature of layered image was extracted. Then, the mean value and variance of low-frequency transform coefficient of the layered image were extracted as the transform feature of the lifting-based wavelet. Finally, all kinds of features were synthesized and used for image retrieval, and the distance measurement formula was defined. The image shape features extracted by the designed algorithm proposed herein could enable the mean value of distances between all standard test images to be 0.2352. The retrieval experiment results on Corel image set are superior to RIM algorithm and FWTH algorithm, which indicates that the proposed algorithm has certain application prospects in the field of image retrieval.