A new approach of two tier image feature vector matching technique using an unsupervised clustering technique

In this paper, we present a new approach to reduce the computational complexity of pattern verification and identification through the process of automatically partitioning the existing image feature vector set into highly relative clusters based on the Optimized Agglomerative Clustering (OAC) technique. The proposed approach consists of two tiers: clustering and matching. In the first tier, the entire trained images feature vector set is automatically divided into distinct number of clusters based on the OAC. Then, the range of centroids minimum and maximum calculate over each individual cluster in the resulting cluster based on the proposed Min-Max Centroid Measure (MMCM). In the second tier, a new matching technique called Two-Level Matching (TLM) is also proposed. TLM aims to match the test image feature vector over the related trained image feature vectors in the specific cluster through two levels. In the first level, TLM verifies the test image feature vector is valid by mapping the centroid of test image feature vector over the range of centroids (Min&Max) of each individual cluster. In the second level, the TLM maps the valid test image feature vector among the related existing image feature vectors in the specific cluster and generates the matching result. The proposed approach reduces the time complexity of image feature vector identification and improves accuracy of the image feature vector matching. The experiment results show that the proposed approach is faster and improves image recognition in both verification and identification scenarios. The related literature survey is described in the next section.