CBMIR System Based on Matrix Weighting Framework and Linear Transformation with KNN

Biomedical images are utilized by radiologists and physicians in cases where diagnosis and recognition would be more efficient. Therefore, it is recommended to propose novel methods that can retrieve medical images with a high percentage of accuracy. A comprehensive feature selection and weighting combination method with novel learning of k-nearest neighbor artificial neural network (KNN-ANN) was introduced for retrieval of biomedical MR images. Modified Radon, and modified Hu moments operators with weighting combinational methods were proposed for achieving a higher percentage of retrieval. Moreover, these characteristics are re-composed for presenting outstanding statistic specification and spatial signals. This spatial and frequency information is obtained for all MR image datasets. The composition of shape and textural features presents robust vectors for retrieval of the biomedical database. In addition, a KNN-ANN framework is proposed and applied to measure the similarity between the query and biomedical database. The presented system achieves 90% retrieval for all types of classes. For matching features, the most similar highest priority principle with KNN is used. The image retrieved from the databases is the image which has less distance and less most-similar highest priority at KNN. The proposed algorithm is evaluated on two different databases. This novel scheme illustrates higher and better specialty in the MRI datasets. The results were compared and understood to be remarkable.

Cite this article as: S. Shafei, H. Vahdati, T. Sedghi and A. Charmin, “CBMIR system based on matrix weighting framework and linear transformation with KNN,” Electrica., 22(2), 258-266, 2022.