In order to improve the retrieval accuracy of images in datasets, an image retrieval algorithm based on weighted distance and multivariate polar harmonic transformation is designed by making full use of the texture and shape features of the query object. The color features are extracted in the Hue-Saturation-Value(HSV) space of the query image. Bessel K-distribution and Non-down Sampled Shearlet Transform(NSST) are introduced to obtain the texture features of the query image for improving its robustness to blur and brightness transformation. With the help of the Quaternion Polar Harmonic Transform(QPHT) mechanism, the QPHT modulus of an image is regarded as a shape feature to improve the robustness to noise and geometric transformation. By fusing the three features, the corresponding feature distance between the query image and the database image is calculated, and the corresponding weight of them is given to measure the similarity so as to output the retrieval results accurately. The test data show that this algorithm has higher retrieval accuracy and robustness, which can still accurately retrieve the target under various geometric transformation attacks compared with the current content-based image retrieval technology.