Quantum Recommendation System for Image Feature Matching and Pattern Recognition

Abstract

The clustering and the classification techniques for pattern recognition are applied in variety of ways, but the problem of clustering binary vectors has not been thorough analyzed. This paper provides a novel approach towards the problem of pattern recognition through the ORB image descriptors. An advanced clustering method is provided in order to deal with the binary image feature descriptors, thus providing the opportunity of adding new classes of recognizable objects later-on: the k-majority algorithm over ORB descriptors is applied, where the Jaccard-Needham dissimilarity measure is used as a distance measure step of the algorithm. It is established, that the following methodology is well suited for a quantum interpretation of the system. A detailed analysis of such transformation is conducted and the Grover’s algorithm is proposed for providing the opportunity to search a specific feature in the available clusters, while reducing the number of iterations of the k-majority routine. In addition to the presentation of the system, described above, this paper provides also the main steps in constructing a similar recommendation system. To that the transformation from a classical to a quantum representation algorithm is described in detail. Such approach can be applied later-on in other applications. Both, the computational complexity and the verification correctness are also indicated below.