I'm currently working on an application that should mark portraits strictly one mark per one portrait. So currently I'm stuck with proper bounding of extracted contours. Here are Canny images. I tried to get bounding rectangles from contours, but then I have too many rectangles and I have to merge/filter them some how to achieve one to one correlation. Also, I tried to approx contours to polygons and filter only polygons with 4 points, but in this case photo frames which are visible on Canny are not approxed to 4 points polygons (i've tried different ranges of epsilon from 1 to 10 percents). I would appreciate for any ideas how come up with those boundaries[C:\fakepath\image.png][C:\fakepath\image-1.png](/upfiles/15167328167189085.png)(/upfiles/1516732795660617.png)