Techniques for enhancing digital images

Abstract

The images obtain from either research studies or optical instruments are often corrupted with noise. Image denoising involves the manipulation of image data to produce a visually high quality image. This thesis reviews the existing denoising algorithms and the filtering approaches available for enhancing images and/or data transmission. Spatial-domain and Transform-domain digital image filtering algorithms have been used in the past to suppress different noise models. The different noise models can be either additive or multiplicative. Selection of the denoising algorithm is application dependent. It is necessary to have knowledge about the noise present in the image so as to select the appropriated denoising algorithm. Noise models may include Gaussian noise, Salt and Pepper noise, Speckle noise and Brownian noise. The Wavelet Transform is similar to the Fourier transform with a completely different merit function. The main difference between Wavelet transform and Fourier transform is that, in the Wavelet Transform, Wavelets are localized in both time and frequency. In the standard Fourier Transform, Wavelets are only localized in frequency. Wavelet analysis consists of breaking up the signal into shifted and scales versions of the original (or mother) Wavelet. The Wiener Filter (mean squared estimation error) finds implementations as a LMS filter (least mean squares), RLS filter (recursive least squares), or Kalman filter. Quantitative measure (metrics) of the comparison of the denoising algorithms is provided by calculating the Peak Signal to Noise Ratio (PSNR), the Mean Square Error (MSE) value and the Mean Absolute Error (MAE) evaluation factors. A combination of metrics including the PSNR, MSE, and MAE are often required to clearly assess the model performance.