A K-Means Based Approach to Efficient Content-Based Image Retrieval

In this paper, we developed a novel CBIR system based on an effective k-means based algorithm. The developed system employs diverse MPEG-7 image feature descriptors for estimating the similarity, which are two color descriptors: CLD and DCD; one texture descriptor: EHD; and one shape descriptor: RSD. The merged features retrieval reports superior retrieval outcomes. So, our proposed CBIR system is employing color, shape, and texture merged features by granting weights for the feature vectors. The introduced k-means based clustering algorithm has been proposed as a preprocessing procedure to accelerate image retrieval and to enhance image retrieval accuracy. The experimental outcomes based on WANG images have been investigated and indicated considerable refinement in terms of average recall and average precision compared with the state-of-art methods.