PERFORMANCE AND QUALITY IMPROVEMENT OF MPEG-4 VIDEO ENCODER

Abstract

Transmission of real-time video over wireless channels is becoming feasible, as high data .rates are available with the advent of third generation mobile communication. The latest multimedia standard MPEG-4 Simple Profile ( SP) is used in Mobile Multimedia Messaging ( MMS ), Video transmission and Packet Switched Streaming (PSS ) applications. MMS and PSS are used for multimedia content delivery over wireless networks. MPEG-4 is ideally suited for wireless channels because of its high coding efficiency, error resilience and content-based functionalities. In mobile solutions, real time encoding efficiency without much computational cost and low-power consumption are essential. Due to the increased complexity of compression tools, MPEG-4 video encoding processes require significant amount of computation and processing power. With the ever-increasing growth in handheld and mobile devices, embedded system applications have become complex and RISC cores are increasingly being used. The RISC processors are used in hand held devices because of their low power consumption, high performance and small in silicon, which are important for mobile devices. Hence, RISC cores are being used to perform video encoding and decoding in mobile video communications. As a result, optimization of applications to the extent possible is essential. Our work is aimed at performing efficient encoding of QCIF (4:2:0 format) video at 15 frames per second, which is sufficient for mobile video communications, on efficient RISC core ( TMS320C64XX DSK ). The video sequence cannot be transmitted as it is since it consumes lot of bandwidth and storage capacity. Motion estimation is implemented at the 'encoder side, which is a very essential block for efficient motion prediction thus reducing the amount of bandwidth and storage capacity that are required. Full search block matching motion estimation is superior to any other Fast block matching motion estimation methods. It does not get trapped onto the local minima while estimating motion. But it consumes lot of computations which is not desired in mobile communication. Our aim in this work is to reduce the amount of computations at iv the same time not compromising on the quality of the video sequence obtained after motion compensation. We have experimented on motion estimation using Fast block motion estimation techniques, motion estimation using multi resolution analysis, motion estimation using spatio-temporal contextual information and combining multi resolution analysis with spatio-temporal prediction to obtain good motion vector prediction and at the same time reduce the computational cost. The quality of the reconstructed video sequence is measured using peak signal to noise ratio (PSNR ).The code is ported on to TMS320C64XX DSK and we tried to optimize the code. v