In a recently published patent application, Samsung disclosed an invention that puts shallow depth-of-field capabilities into tiny, point and shoot cameras. Samsung gets there by using dual lenses in a single camera and using that second lens in a rather novel way. While the primary lens is capturing the full-resolution image, a secondary lens and sensor captures another image with the sole purpose of evaluating the relative distances of areas in the image. (In some configurations, the camera may actually be a 3D camera, while in other configurations, it may truly be a secondary lens and sensor.) Then, the camera merges the data with the primary image to create a depth map and applies a graduated blur based on this depth map.
That's a horrible sample pair. The bokeh looks unnatural to me, particularly how objects closer than the baby's head are sharp. Hopefully, they can improve the algorhythms.