Near-duplicate image recognition based on the rank distribution of the brightness clusters cardinality

In this paper the usage of multi-step segmentation for near-duplicate image recognition is investigated. The clustering of image pixels brightness is used for segmentation. The clustering is realized by means of a recurrent neural network.
The search pattern based on the rank distributions of the brightness clusters cardinality is suggested. Experimental results on the near-duplicate image recognition based on the application of the suggested search pattern are given. It is shown that the use of a multi-step segmentation and rank distributions of the brightness clusters cardinality allows one to successfully recognize the duplicates, which are received by a considerable visual distortion of the original image or by the change of image scale.