Implementing a Low-Complexity, Adaptive, Layered Video Coder for Video Teleconferencing

摘要

Real-time interactive video applications, such as video teleconferencing, present difficult challenges to network designers due to strict quality of service constraints and the limitations of traditional video compression schemes. These limitations reveal themselves notably in two areas: poor error robustness and a lack of flexibility when dealing with multicast scenarios over heterogeneous networks. A more promising approach that improves error robustness while also offering a solution to the network heterogeneity problem is to employ a layered video codec. This thesis presents the implementation of a new layered video coder scheme. Block updating coupled with an aging algorithm is used in this scheme to select macroblocks for transmission. Block updating selects macroblocks that have changed due to scene motion, and the aging algorithm ensures that an entire frame is transmitted within a set time interval. Layering is accomplished through application of the fast Haar transform and/or the discrete cosine transform. Layer assignments are made by grouping bands of coefficients with similar variances. Quantization and encoding for motion video employs both an industry standard and uniform quantization with a custom variable length coding table. For static slides, uniform quantization and a second custom variable length coding table are employed. Rate control is accomplished via the reduction of a four-dimensional operational distortion surface to a one-dimensional optimal curve implemented as a simple table lookup of quantizers.