MSU Scene Change Detector (SCD)
- Project, Ideas: Dr. Dmitriy Vatolin, Alexander Parshin
- Implementation: Ivan Glazistov
- Updating and additions: Sergey Grishin
Scene Change Detector is made to automatic identification of scene boundaries in video sequence.
[!] — Known bug
[+] — New Feature
[*] — Other
[*] Windows Vista & Windows 7 support implemented
[*] Visualization bug fixed for non-stadard resolution video
[+] First plugin release
The plugin implements four algorithms of similarity measurements between two adjacency frames in video sequence:
- Pixel-level frames comparison
- Global Histogram comparison
- Block-Based Histogram comparison
- Motion-Based similarity measure
The choice of the algorithm can be made in Settings. Numbers from 1 up to 4 corresponds to each algorithm.
Default and recommended value is 3 (Block-Based Histogram).
Y-plane is drawing during the visualization. Brightness of scene boundary frames is increased.
Example of visualization:
Metric’s plot is making after all measurements. “One” value means that current frame is the first frame in scene, other frames have “zero” values. Sequence average value is the number of detected scene changes.
Similarity measure of two frames is the sum of absolute differences (SAD) between corresponding pixels values.
The histogram is obtained by counting the number of pixels in frame with specified brightness level. The difference between two histograms is then determined calculating SAD of number of pixels on each brightness level.
Each frame is divided into 16x16 pixel blocks. Brightness distribution histogram is constructed for each block. Then similarity measure for each block is obtained. Average value of these measures is accepted as a frames similarity measure.
Motion Estimation algorithm with block size 16x16 pixels is performed for two adjacency frames at the first stage. After that average value of motion vector errors is accepted as a finally similarity measure.
- Codecs Comparison & Optimization
- Video Filters
- Video Quality Measurement Tool 3D
Semiautomatic Visual-Attention Modeling
MSU Benchmark Collection