Video comparison Python project

The algorithm we implemented can be described as follows:

• ⁠Preprocessing (all information about the videos is extracted, aspect ratio, resolution and fps are taken into account for detailing further analysis);
• Video frame generators are created based on derived parameters;
• To each pair of frames the background subtraction is applied sequentially (clahe for further stabilization effect + bg sub);
• The result of bg sub is followed with some morphological operations to eliminate small noise and improve strong regions of difference;
• Then we search for rectangular regions of difference on resulting mask and apply SSIM check (and, if SSIM indicates difference indeed - additional SIFT check) to each region;
• Once a number of different pairs reach the preset threshold of all frames or a big portion of consequent frames is different (running window approach), videos are labeled as different. If the threshold is not reached until the end, videos are considered equal.