Enhancing the Quality of Melanoma Dermatoscopic Images Using Wavelet Coefficients within a Deep Learning Framework

نویسندگان
doi
10.61882/aswtr.22.2.48
چکیده

Background: High-quality medical images improve accuracy in diagnosing diseases. Melanoma is a common and deadly skin cancer. Dermatoscopes offer a non-invasive way to capture skin lesions, but dermatoscopic image quality greatly affects diagnosis. Methods: We propose a deep learning–based method to enhance dermatoscopic image quality using wavelet detail coefficients. ASuper-Resolution Convolutional Neural Network (SRCNN) estimates high-resolution wavelet coefficients from low-resolution images. This approach enables efficient training with fewer samples and lower computational cost. Refinement of these coefficients leads to better image reconstruction, measured using Peak Signal-to-Noise Ratio (PSNR). Results: Tests on 180 dermatoscopic images show our method achieves a PSNR of 48.99. Previous approaches reached a maximum PSNR of 44.02. Conclusion: This result shows our framework can provide high-quality images, supporting more accurate melanoma diagnosis.