# Clustering algorithm for image metric

Im working on image clustering (finding duplicates). I have a metric for images, it uses histogram features (mean, dispersion, skewnewss) for each color channel. So there are 9 dimensions. It is quite easy to calculate distance between two images (L1), but I don't know how to cluster them. Is there any clustering algorithms that use only the metric, other than bruteforce? Wikipedia article tells about projected clustering, how can it be implemented?

• Yes, there are algorithms using only metric, but they must require to compute the distance between almost each pair of images in average, because that's the only way to identify two images that are close to each other when most other pairs are far, and this can make the algorithm very time consuming if there are many images. I'll try to find an article later. Commented Sep 25, 2015 at 17:22
• However, if you already have features in 9 dimensions, then it's better to use those features directly for clustering than a method that uses only the metric. Commented Sep 25, 2015 at 17:22
• This might be more suitable for Cross Validated. Commented Sep 25, 2015 at 22:42
• @ZsbánAmbrus, thanks for help. How may look the algorithm which is using only metric? Should it build graphs or something? Because there is the problem how to treat three images ABC if distance between AB and BC is short but AC is far. Commented Sep 26, 2015 at 17:05