| dc.description.abstract | The video coding technique has been extensively applied in many scenarios of our daily
life. For example: vehicle electronics applications, mobile phone, computer, high-definition
television ...etc. To support such various applications, a versatile video coding scheme is
essential. Therefore, ITU-T Video Coding Experts Group has developed a video standard
H.264/SVC [1], an extension of H.264/AVC [7]. It can provide the bitstream adaption to fit in
with different resolution, network condition and hardware capability. The H.264/SVC
includes three types of scalability: temporal, spatial and SNR scalability. However, the
complexity of H.264/SVC decoder is very high. For this reason, how to reduce the complexity
of H.264/SVC is very important issue.
In this thesis, we realize the H.264/SVC decoder with Texas Instrument DM6437 DSP
platform. This work transplants the reference software JSVM9.16 [2] to the DSP platform
DM6437 [3]. We separately analyze the complexity of H.264/SVC decoder in for temporal,
spatial, SNR and Combine scalability. According to the analysis result, we optimize each
module of H.264/SVC decoder. This thesis proposed two optimization methodologies in
inverse discrete cosine transform (IDCT) module and up-sampling module. With both
optimizations, the performance of IDCT and up sampling module can be increased as high as
10.7x and 12.29x, respectively. Besides, we also utilized the Code Composer Studio (CCS) to
draw up the memory mapping and explored some special intrinsic instructions to improve the
decoding performance. The overall decoder system can speed up 5.79x on average.
Finally, we will show a complete H.264/SVC decoder system. It is implemented in
DM6437 DSK. This system received the bitstream through network and decoded the
bitstream to obtain video information. Furthermore, the bitstream extractor is also
transplanted into this system. The decoded video information is displayed on LCD monitor.
To switch scalability function, we applied DIP switchs to separately enable temporal, spatial,
SNR, and combine scalability.
| en_US |