Low Complexity H.264 Encoder Using Machine Learning For Streaming Applications

H.264, MPEG-4 part-10 or AVC, is the latest digital video codec standard which has proven to be superior than earlier standards in terms of compression ratio, quality,bit rates and error resilience [1]. Joint model (JM) reference software is used for academic reference and it was developed by the Joint Video Team (JVT) of ISO/IECMPEG and ITU-T VCEG (Video coding experts group). The Intel IPP H.264 (Integrated Performance Primitives) is a product of Intel which uses Intel IPP libraries and SIMD instructions available on modern processors. The Intel IPP H.264 is multithreaded and uses CPU optimized IPP routines. These two softwares are compared interms of execution time and video quality of the decoded sequences. The metrics used for comparison are SSIM (Structural Similarity Index Metric), PSNR (Peak-to-Peak Signal to Noise Ratio), MSE (Mean Square Error), motion estimation time, encoding time, decoding time and the compression ratio of the H.264 file size (encoded output). The compression ratio of H.264 file is found to be less in JM software at various bit rates than in Intel IPP. Hence, it is preferred over Intel IPP for reduction in the motion estimation time during encoding.Motion estimation takes about 60 to 70 percent of the encoding time. The time consuming Sum of Absolute Dierences (SAD) method adopted in the H.264 encoder in JM 16.2 software is replaced with a classication rule using machine learning. This tree is implemented in the form of if-else statements in the motion estimation block ofJM16.2. Hence, the motion estimation process is reduced to if else statements thereby reducing the encoding time. H.264 is a video codec format. Its corresponding .AAC (Advanced Audio Coding) audio format and the video format are then placed on a MP4 container using an open source tool called MP4box. This MP4 le can be streamed (after forming manifest files) using IIS (Internet Information Services) to achieve smooth low complexity streaming of media over the Internet.