0.7 Adaptive roi results

Implementation with a widescreen movie

To evaluate our adaptive ROI system, a scene from the movie “Punch Drunk Love” was analyzed. A straight-cut fullscreen version was generated, and our adaptive ROI system was used to generate a pan-and-scan fullscreen version. The pan-and-scan version was analyzed by tracking the ROI midpoint as a function of frame number.

Roi midpoint tracking

To illustrate the mobility of the ROI midpoint, the following charts show the deviation of the midpoint from center after processing by motion, edge, and focus detection individually and combined. The first 4 charts show the unfiltered results of processing, while the last chart shows the filtered ROI midpoint using the combined processing.

Results of processing: motion only

Results of processing: edges only

Results of processing: focus only

Results of processing: combined methods, no filtering

Results of processing: combined methods, with filtering

Image results

The following images illustrate the functionality of our adaptive ROI system compared to a center-cut fullscreen image. The images are taken from the same scene as was used for the above analysis, in the Frame # 1000 to 2000 range.

Original widescreen frame

Result of center-cut fullscreen conversion

Result of adaptive roi fullscreen conversion

Video Results

Here you can download a center-cut fullscreen and an Adaptive ROI fullscreen version of the scene above. The videos “with black bars” show a 16:9 video with black bars over the regions that would be cropped out – the ROI is still 4:3.

Note that the videos go further in the scene than the charts above.

Adaptive ROI fullscreen

Adaptive ROI with black bars

Center-cut fullscreen

Center-cut with black bars