Improvement of surveillance video coding based on IEEE 1857 standard

IEEE 1857 Video Coding standard is introduced with the aims to double compression rate for high efficiently and low complexity for video surveillances. However, to compress with high compression ratio is affected to recognition. In this research, we propose a new algorithm to improve SIFT similarity by using feature map, which is detected keypoint features from SIFT detection. Based on the number of keypoints, macroblock (MB) of frames are divided into two groups: the MB which has more features uses as an important MB group and others for non-important MB group. If the current MB is in important MB group, that MB is encoded with lower quantization parameter (QP) so the features in that MB can preserve after compression. For other MBs, the high QP is used so the overall performance is maintained. On the experiment, the effectiveness of the proposed algorithm is evaluated by comparing the SIFT similarity values between current reference software of IEEE 1857 standard with disable background modeling and proposed algorithm. According to the experimental results, our proposed algorithm can improve the SIFT Similarity values up to 49% when encoded with high target bits and Low QP in the specific important MBs based on to improve the MBs feature preservation, which is shown that the proposed algorithm outperforms to the current reference software in terms of improve SIFT similarity while maintain the same PSNR and bitrate values.