Improved binary codes for efficient content-based image retrieval

The usage of high-dimensional features to encode the image data shows comparative advantages with large-scale image processing problems, which may cause the "Curse of Dimensionality". To deal with this case, hashing methods have achieved great interests in content-based image retrieval. Here we concentrate on how to figure out efficient search via employing binary codes. The key point of our proposed method is to get the optimized mapping from the feature space to the Hamming space to get the improved binary codes and accelerate search. Our method consists of two parts: firstly we generate a projection matrix to reduce the dimensionality and raise the separability of features in derived subspace; secondly a rotation matrix is used to minimize the quantization loss. Experimental results on two classic databases testify that our proposed scheme outperforms several state-of-the-art methods.