Low complexity AVS-M using Machine learning algorithm C4.5

Macroblock mode decision is the most expensive process in terms of computational power required. In any video codec motion estimation along with the macroblock mode decision consumes approximately 80% of the encoding time, resulting in encoding maximum of only 2 frames per second. This makes it almost impossible to implement a video codec without using specialized hardware, which causes problems like power consumption and overheating of device in low end devices like mobile, and notebooks. An effort is made here to reduce the encoding time, by implementing Machine learning algorithm C4.5, in the block decision block. The proposed encoder, on an average reduces the encoding time of the sequence by 75%, with an average loss of only 2% in PSNR while saving considerable number of bits used to encode the sequence.