Image information retrieval plays a significant role in many application areas, such as satellite image databases, geographic information systems, diagnostic medical image databases, and commercial logo databases. Hence, developing an effective and efficient image retrieval system is very important. This paper proposed an image retrieval system based on the VQ (vector quantization) index and spatial locations of color-mass-center in images. VQ is used to construct a lower-resolution version of the original color histograms of images. The images in the database are divided into clusters based on VQ code index. This system describes the color-spatial feature of an image using the relative location of four color-mass-centers. The color-mass of a pixel is defined by the gray level of the pixel. We divide an image into four quadrants. Each quadrant has a color-mass-center. The Euclidean distances between each pair of four color-mass-centers are the indexing key to search the image. Experimental results show that the recognition ability is insensitive to the shift, rotation, and scale variants of images.
