A Comparative Analysis of Denoising Filters on Thalassemia Images in Red Blood Cells

Thalassemia is a health problem in human red blood cells, and the number of cases continues to increase. Medically, the diagnosis of this disease is done manually. Current medical technology development has reached the use of artificial intelligence techniques because the output of medical devices is in the form of digital signals and images. However, these techniques have limitations due to noise in the image to be processed. Thus, an effective noise reduction technique is needed. This research aims to evaluate the best filter to improve image quality and support accurate diagnosis results. Three filters are implemented for the noisy thalassemia image. The simulation is required to test the performance of filters by adding several noise types (i.e., Gaussian, Mean, and Median filters). Each filter's performance is evaluated using PSNR, SNR, MSE, and SSIM metrics. The results show that the Median filter can effectively remove low-level noise with Gaussian noise, while the Mean filter can remove high-level noise. On salt & pepper noise, it shows that the median filter performs best on all noise levels from low to high. The results show that with Gaussian noise, the Median filter can effectively remove low-level noise with an average PSNR improvement of 36,95 dB. In comparison, the Mean filter can remove high-level noise with an average PSNR improvement of 29.10 dB for level 2, 20,47 dB for level 3, and 18,37 dB for level 4. On salt & pepper noise, the median filter performs best on all noise levels, especially at low noise levels with an average PSNR value of 47.20 dB.