Performance analysis of video magnification techniques in perceiving of subtle small motions

In this thesis, performance analysis of the video magnification techniques which are used to reveal subtle variations of colour and motions in a video that are difficult to perceive with the human eye was achieved. In this analysis process, Eulerian colour magnification and Eulerian motion magnification techniques have been used as video magnification technique. In both video magnification techniques, both Laplace pyramid and Gauss pyramid have been used as video decomposition tecnique and performances of these pyramids have been compared. In Eulerian Video Magnification technique, the input video is firstly decomposed into different spatial frequency bands by using a pyramid structure. After decomposition process into these spatial frequency bands, the same temporal filter is applied to all bands. Filtered spatial bands are magnified by a magnification factor that is specified by the user. These magnified signals are then added back to the original signal. All of these video frames that are decomposed as spatial are reconstituted to obtain the final output video that contains magnified colour and motion. The output videos that were obtained from the all pyramid decompositions are computationally analyzed and compared with each other through PSNR(Peak Signal-to- Noise Ratio) and SSIM(Structural Similarity Index Measurement) values. The video processing time is a parameter which is used to compare with video decomposition techniques. When the results of the experiments that are done with video decomposition tecniques investigated in terms of video processing time, it was seen that video magnification with Gauss pyramid decomposition has a better performance. When these techniques investigated in terms of PSNR values as video quality, it was seen that while Gauss pyramid gives a better result in colour magnification process, Laplace pyramid gives a beter result in motion magnification process.