An optimal self adaptive deep neural network and spine‐kernelled chirplet transform for image registration

• Published in: Concurrency and computation Vol. 34; no. 27
• Main Authors: Senthil Pandi, S., Senthilselvi, A., Maragatharajan, M., Manju, I.
• Format: Journal Article
• Language: English
• Published: Hoboken, USA John Wiley & Sons, Inc 10.12.2022; Wiley Subscription Services, Inc
• Subjects: Accuracy; Artificial neural networks; dynamic group; Image enhancement; Image processing; Image registration; Medical imaging; modified spine‐kernelled chirplet transform; Neural networks; Registration; self‐adaptive deep neural network; Uncertainty analysis
• ISSN: 1532-0626; 1532-0634
• DOI: 10.1002/cpe.7297

Summary Image registration is one of the image processing techniques that align more than two images of a similar scene captured under different perspectives at different intervals of time. In demographical research, the image registration process assists to study differences in the structure of brain tissue. Due to enhanced technological advancements, numerous image registration methods have been established. On the other hand, these traditional techniques face few real‐time challenges while processing huge input data. In addition to this, uncertainty analysis becomes a crucial step in medical applications which is utilized to judge whether the registration result is valuable or not. The high percentage of uncertainty than the threshold makes the registration result abnormal. Therefore, to conquer such circumstances, this research work proposed a modified spine‐kernelled chirplet transform (MCST) based optimal Self‐Adaptive Deep Neural Network (SADNN) which focuses mainly on enhancing registration accuracy by reducing the uncertainties of registration results. The experimental analysis is conducted and from the evaluation results, the proposed MCST‐based optimal SADNN technique outperforms existing techniques in terms of accuracy, specificity, sensitivity, F‐measure, and DICE values. Moreover, the proposed method achieves 97.2% accuracy for accurate image registration.