Robust method for the transmission of DCT coded images and image quality evaluation of the received images

Multimedia communication in real time requires fast and reliable transmission of large amounts of data. Due to the limited availability of channel bandwidth, the multimedia data are compressed before transmission. A widely used compression algorithm is based on the discrete cosine transform (DCT) which has been adopted in the JPEG standard. However, compressed data are extremely susceptible to channel noise and error. This dissertation presents a novel error detection and correction methodology for corrupted DCT coefficients caused by transmission of DCT coded images over a noisy channel. This method is based on the orthogonal property of discrete cosine transforms and the transmission of actual intensity levels of a few reference pixels. It is possible to correct t corrupted DCT coefficients in an image block with the knowledge of 2t + 1 pixel intensity levels for each image block. The transmission of the reference pixel intensity levels can be avoided by replacing the reference pixel intensities of each image block by a known intensity level prior to transmission. When the number of corrupted DCT coefficients exceeds the correction capability of the proposed detection and correction algorithm, we employ an error concealment technique to recover the corrupted block. Results indicate that this algorithm provides an excellent balance between correction capability and channel overhead requirement. The second part of this dissertation presents a distortion measure for images that have undergone transmission and error concealment. The peak signal to noise ratio (PSNR) is widely used in evaluating image quality. However, PSNR is a statistical measure that does not resemble the image quality evaluation process employed by humans since it weighs all the errors equally while the eyes react differently to different error patterns. We present a distortion measure that treats an image as a collection of several blocks and takes into account the characteristics of the human visual system. By treating the image as a collection of blocks, the algorithm to determine image quality is more accurate in evaluating local distortion and providing a measure of the extent of distortion in the image. Results indicate that the algorithm resembles the human evaluation of images much more closely than the PSNR