A New Method for Reverse Prediction of MPEG frames in Video Streaming

Recently macroblock based technique for backward playback of MPEG video streaming has been discussed in which the network bandwidth and buffer size requirements are less. In this technique, the macroblocks are divided into two categories backward macroblock (BMB) and forward/backward macroblock (FBMB) based upon their motion characteristics. In this paper, we propose a new method in which the network bandwidth requirement further reduces without incurring cost on the buffer. We predict the required I or P-frame from the currently decoded frame which resides in the frame buffer at the client system. The motion vector information of the current frame is used to find the position of its various pixels in the previous frame and the prediction errors of the current frame are used to find their exact values. We performed extensive simulations on various videos having slow, medium and high motion characteristics. Experimental results show that on an average 93.4% of pixels in previous I- or P-frame can be reverse-predicted from the current P-frame. This makes considerable saving in the network bandwidth requirement as fewer pixels are requested from the server.