I have two photos that were taken in a burst (the movement between the two is very small, exploiting the hand jitters between the two images), and I have to use the two to suppress the background and separate the wanted object from the background. The user of the app presses on the object to separate.
I tried using SGBM to find the depth map, and then I could threshold the values close to the values in the point the user pressed, but the results are horrible (no matter the parameters chosen - my guess is that is because the images are not exactly left-right versions of the same scene, but slightly lowered and heightened as well).
What is the best course of action to do what I need? I have to use a pair of very close images, which are randomly tilted from each other. Is the fact that the images are so close to each other problematic? I tried everything I can but nothing works.