Caption processing for MPEG video in MC-DCT compressed domain

Manipulating lossless video in the compressed domain

“…Most of the previous work on mapping algorithms into the compressed domain has focused on formats such as JPEG that utilize a Discrete Cosine Transform (DCT) to achieve spatial compression [1,5,6,8,14,15,19,20,23,27]. This task requires a different analysis, with particular attention given to details such as the blocked decomposition of the image, quantization of DCT coefficients, zig-zag ordering, and so-on.…”

“…total volume of data processed, thereby offering large savings in both execution time and memory footprint. However, existing techniques for operating directly on compressed data are largely limited to lossy compression formats such as JPEG [6,8,14,19,20,23] and MPEG [1,5,15,27,28]. While these formats are used pervasively in the distribution of image and video content, they are rarely used during the production of such content.…”

Internet-based interactive HDTV

“…This way, the motion compensation work for reference macroblocks is eliminated since we only need to know how they have changed and can adjust the prediction errors based on the changes calculated using only the watermark embedded. Subsequent to this, Nang et al [10] pointed out that [12]'s approach may not work: if the embedded watermark or captions are too strong (luminance values are too large), Eq. 1 will make E ij overflow the [16,235] bound for a pixel's luminance value.…”

“…Typical inverse motion prediction link discovered by backtracking prevent overflow, the maximum luminance value of the reference macroblock is needed again and the savings of [12] are no longer available. The authors of [10] avoided this problem heuristically by using the average luminance of the background block to estimate the caption threshold, which is the DC coefficient in the DCT domain and can be easily acquired. The potential problem is that, when the maximum pixel value differs significantly from the average, overflow will still occur, but the authors claim that the resulting image is of acceptable quality.…”

“…In summary, for general-purpose video/image/text embedding, we must have the original macroblock value of both the reference macroblocks and the dependent macroblocks in the RD domain via motion compensation and so cannot avoid it as in [10,12]. Therefore, we have to face the computationintensive motion compensation operation and find a new solution to minimize the amount of computation involved in motion compensation.…”