Image Postprocessing Technologies In DCT Domain

With the rapid development of the internet and mobile communications, multimedia information delivery becomes more and more popular. For reducing the data amount and meeting with real-time restriction, images and video compression is certainly the key technology in multimedia communications. Image resizing in DCT domain will help us to transmit images while the bandwidth is limited. Lots of compression tools had been put into effect, including JPEG, MPEG-x, and H.26x. These compression standards effectively compress the image and video data, and they adopt the"block DCT + quantization"architecture. Consequently, distortions are introduced into the reconstructed images, such as blocking artifacts, mosquito effect, ringing, staircase effect, etc. When the video is compressed under low bite rate, the blocking effect is the major factor affecting image quality. Many de-blocking methods have been proposed to tackle it. Unfortunately, the existing methods usually suppress the details during deblocking. For most digital consumer electronic equipments, stored image and video can be enhanced directly in DCT domain. Therefore, de-blocking in DCT domain could improve the computational efficiency. We also include human visual system for image quality assessment in this paperIn this dissertation, much work is proposed for video post/pre-processing in DCT domain including image enhancement, de-blocking and image resizing.For the reason of blocking artifacts and its distribution on DCT coefficients, we propose a new POCS based de-blocking method, called PBOFN (POCS Based Overlapped Four Neighbor blocks de-block algorithm), in DCT domain. A new smoothness constraint set is proposed to process the combination of overlapped four neighbor blocks and the original block in turn. The smoothness constraint set and an adaptive quantize constraint set are cooperated to smooth the blocking artifacts. We can acquire the best result only after once iteration. The experiments on four standard images show that our proposed method is superior to other five control methods in all aspects, such as PSNR, the subjective quality and the MGBIM (the assessment metric of blocking artifacts)In the follows, we study the image enhancement based on BDCT. First, we modify the Tang's method in two ways, named MST and MFT respectively to further improve de-blocking performance. To enhance the image and surpress the blocking artifacts, DCT blocks are cataloged into one of these kinds: smooth block, edge block, texture block and weak texture-smooth block. With considering about the human visual system and the neighbor blocks' properties, different kind blocks are processed in their own ways are also considered, which is called PE (Perceptual Enhancement). Because the background luminances are not adjustable in PE, we combine the PE with background luminance enhancement to form a new method named as PEB. PE and PEB improve the objective quality of the processed image greatly. The EME and MGBIM are also adopted to evaluate the deblocking efficiency.Finally, we study image resizing in DCT domain. Different size block has different resizing performance. A new adaptive image resizing, named as ABSC method, is proposed to adaptively choose block size to get better performance.