Image restoration using PDE and segmentation techniques

Image enhancement as well as restoration methods for remotely sensed data have been widely studied and investigated among researchers who interested in image processing. The main goal of this dissertation is to develop innovative approaches to enhance the image quality and mitigate the probable noises which may infect the remote sensing datasets. In this thesis, we apply Partial Differential Equations (PDE) to efficiently denoise images, while their important details such as edges and boundaries are preserved. The parameters of proposed PDE-based algorithm are adaptively set regarding the level and characteristics of the noise as well as the nature of noisy datasets. Implementing higher order PDEs and developing complex PDEs to gain better results are also presented in this thesis. Moreover, we propose to apply segmentation as a pre-processing step to localize the denoising process. Indeed after extraction of segments, the proposed PDE-based denoising schema is adaptively applied to each segment instead of the whole image. The performance of the presented image restoration techniques is quantitatively measured by different criteria such as Mean Square Error (MSE), Signal to Noise Ratio (SNR), Figure of Merit (FOM), and Structural Similarity (SSIM). For further assessment of the gained results, we also propose to feed the original/noisy/denoised images into a supervised classifier and explore the outcomes. To this end, two well-known classifiers named as Maximum Likelihood (ML) and Support Vector Machine (SVM) are utilized in this work. Extensive set of implementations have been done in MATLAB. The gained results which are drawn in figures and tables imply the efficiency of proposed PDE-based image enhancement method compared to several other well-known existing image noise reduction approaches.