Integrating color into the local features based on the stable color invariant regions for image retrieval

The local features of the objects are widely used in image matching and retrieval. The popular algorithms of the local feature detection are designed for gray images, which failed to make use of the color information in color images. However, the color information is absolutely nontrivial in the discrimination of different objects, vacancy of which would lead to miss-judgment. Although, some color invariants include a lot of color information, the local feature information has not been involved in the color invariants. The simple and efficient method to enhance the performance of the descriptors is to combine color with the local features. Generalized color moment and SIFT are chosen to make this combination in this paper. Because the calculation of the generalized color moment invariants is based on the regions in color image, the stable color invariant regions are first extracted through analyzing the reflection model of the object in the color image. Then, color and local features are fused to establish the framework of image retrieval based on color invariant regions. The robustness of our algorithm is strengthened and the time-consumption in matching is reduced compared to other algorithms.