Blind Image Restoration by using PCA-Subspace Generation and Image Quality Optimization
Image blurring process is commonly formulated as two-dimensional convolution between the latent image and the blurring system. Blind image restoration problem is to estimate the latent image only from the blurred image. Conventional blind image restoration techniques tend to solve this problem by estimating the blurring system and therefore their effectiveness are dependent to the accuracy of their estimation. Principal Component Analysis (PCA)-based restoration technique, however, do not employ PSF estimation and still gives high restoration quality. PCA-based techniques have two different roots. The first is to boost the high-frequency component lost during the blurring process by maximizing the image variance. The other comes from source-separation using PCA. Previously we proposed a PCA-generated subspace for blind restoration and proved its superiority to conventional methods. However, the algorithm should be improved. This study proposes a sign-determination method and modify the image quality optimization technique. The noise robustness and the application to other blurs are also investigated in this paper. From the experiments, we can see that the proposed method gives higher restoration quality both for simulated blur images and real images than conventional methods.
Keywords—Blind image restoration; Single image restoration; Principal Component Analysis; Image enhancement; Image quality