Texture image retrieval based on Log-Polar transform and association rules mining

The extraction technology of rotation, scale, and translation-invariant features is an important research direction in image retrieval. By introducing the data mining technologies into the image retrieval domain and combining the Log-Polar transform, a novel texture image retrieval method based on the association rules mining is proposed in this paper. All the images with rotation, scale and translation distortion are firstly converted to Log-Polar coordinates, then the association rules of the transformed edge images are mined in Log-Polar coordinates by using Apriori algorithm, and at last the association rules similarity is calculated between images by employing the similarity index that is qualitatively consistent with the human visual system. The practical experiment with the Brodatz image database indicates that the presented method is well resistance to the geometric distortion, can achieve high precision for the images with rotation, scale and translation, and shows clear advantages comparing to the traditional ones.