The Reconstruction System For Highly Compressed Image

Image compression always plays an important role in the field of image communication. In the case of the low bit rate, annoying compression artifacts inevitably occur and seriously degraded image quality. The representative compression artifacts are blocking ones which appear as grid noise along the block boundaries because each block is transformed and quantized independently. Thus, the blocking artifact removal is required to achieve high-quality reconstruction from compressed images. In this thesis, we aim to remove annoying blocking artifacts and reconstruct natural-looking results from highly compressed images. The main contributions are described as follows.First, we propose an image de-blocking algorithm based on four separate modes. We have separated the different blocks into four basic modes such as smooth, intermediate, complex, and corner ones by calculating the active value of the block boundaries between two blocks, respectively. Experimental results demonstrate that our method is very effective in blocking artifact removal especially for highly compressed images.Second, we provide a novel generic prior for image enhancement, i.e., the gradient profile prior, which implies the prior knowledge of natural image gradients. In this prior, the image gradients are represented by1-D profiles of gradient magnitudes perpendicular to image structures. The image enhancement can be used to reduce the over-smoothing problem and ringing artifact. Experimental results show that the sharpness of the images is obviously improved and reduces the over-smoothing effects caused by the separate mode-based de-blocking.Third, we present a reconstruction system of highly compressed image based on the three steps of image de-blocking, enhancement, and super-resolution. By the reconstruction system, compression artifacts such as blocking and ringing are effectively removed from compressed images. In particular, the image gradient enhancement method can recover the image details due to both super-resolution and de-blocking. Experimental results are provided to demonstrate the effectiveness of the reconstruction system.