In order to solve the problem that locality preserving projection hashing can results in the poor expression of image feature and lower retrieval efficiency when it is applied in image retrieval, a novel image retrieval method based on hashing combining principal component analysis (PCA) with locality preserving projection (LPP) is proposed. Firstly, the sample is reduced dimension with PCA, a random matrix is introduced to make rotation of the PCA transformational matrix. The original sample is projected into the PCA transformational matrix and the reduced-dimension PCA sample is achieved. Meanwhile, the similarity structure between samples is taken into account. The reduced-sample is mapped with LPP. On these basis, the projection matrix is constructed with a random matrix. Finally, the original sample is projected into the projection matrix and the hash coding is achieved. The presented method can keep the local and overall similarity structure of the samples because of the application of PCA and LPP. Furthermore, the quantization error between codes is reduced by introducing of random rotation, thus improving the efficiency of image retrieval. Experiments show that the proposed method can achieve better performance compared with other traditional methods.
