Evaluating the Impact of a Digital Filter–Bicubic Spline Hybrid on Classifier Performance in Dermoscopic Image Processing

In this research work, a hybrid algorithm combining digital filters and splines was developed to improve the quality of medical images. Digital filters effectively reduce noise while preserving sharp edges and patterns in images. Splines have advantages in filtering and smoothing input images, providing high accuracy in processing complex data. This hybrid technique, which combines the most effective elements of both approaches, allows for enhanced noise reduction and preservation of important information in medical image processing. This study also presents the theoretical basis of hybrid filtering and creates a mathematical model to support it. The "Bilateral + Bicubic" hybrid technique, which combines Bilateral filter and Bicubic interpolation to provide the best filtering result. The ISIC 2019 skin cancer dermoscopic image dataset was used to test the proposed methods. The proposed hybrid method provided filtering with PSNR value of 36.2 dB and MSE of 0.00023. To validate the developed Bilateral + Bicubic approach, dermoscopic images processed based on the hybrid method were fed to ResNet-50 classifier. The overall accuracy of the multi-class classification results of skin cancer improved by 4.7 % based on the proposed approach. The proposed hybrid approach to preprocessing medical images provides the best structural similarity and filtering results, and the model has excellent accuracy and generalization ability across different skin lesion classes. This provides a reliable basis for early and accurate diagnosis of skin cancer in clinical practice. The research paper also discusses the effectiveness of the hybrid algorithm in terms of its application.

Ҳали таржима қилинмаган