A New Feature Set For Content Based Image Retrieval

An efficient feature set for retrieving color images from the image database is proposed in this paper. Features chosen here play an important role in bringing back the images from huge databases. This paper proposes a unique color feature based on vector quantization and texture feature based on the pattern of traversal. The Dominant Codebook (DC) is the new color feature designed from the compressed data achieved from the vector quantization. This Dominant Codebook describes set of pairs of Code Vectors and their percentage occupation in the image. The Scan Pattern Co-occurrence Matrix (SPCM) is the new texture feature designed to capture the traversal of adjacent pixels in a scan pattern. In addition to them, the difference between pixels which present in a particular scan pattern is also considered as another texture feature, this feature is named as Scan Pattern Internal Pixel Difference (SPIPD). These new color and texture features are combined to improve the performance of the image retrieval. The results of the experiments demonstrate that the proposed feature set outperforms the existing methods.