The present paper proposes a new technique that extracts significant structural, texture and local edge features from images. The local features are extracted by a steady local edge response that can sustain the presence of noise, illumination changes. The local edge response image is converted in to a ternary pattern image based on a local threshold. The structural features are derived by extracting shapes in the form of textons. The texture features are derived by constructing grey level co-occurrence matrix (GLCM) on the derived texton image. A new variant of K-means clustering scheme is proposed for clustering of images. The proposed method is compared with various methods of image retrieval based on data mining techniques. The experimental results on Wang dataset shows the efficacy of the proposed method over the other methods.