Shape from defocus Basic idea of shape from

Shape from defocus • Basic idea of shape from defocus • If a point is located on the focal plane, then all the rays (green lines) from this point will converge to a single sensor point and the image will appear sharp • If a point is not located on the focal plane, then all the rays (blue lines) will reach multiple image sensor points and result in a blurred image (the blurred region is called circle of confusion)

Shape from defocus • Basic idea of shape from defocus • If a point is not located on the focal plane, then all the rays (blue lines) will reach multiple image sensor points and result in a blurred image (the blurred region is called circle of confusion) • The diameter of the circle of confusion is:

Shape from defocus • Basic idea of shape from defocus • • c can be measured by the width of an edge

Shape from defocus • Basic idea of shape from defocus • How to mathematically measure blurness

Shape from defocus • Example

Shape from defocus • Let’s apply the state-of-the-art shape from defocus method on MIT dataset

Shape from defocus • We can see that shape from defocus is difficult to detect subtle depth change

Shape from defocus • Apply the state-of-the-art shape from defocus method on NYU dataset

Shape from defocus • It looks like images in NYU dataset are not strictly focused at some point

Shape from defocus • Successful cases:

Shape from defocus • The preliminary results show that • The shape from defocus method works when there are significant variations of blurriness in the image • When depth change is small, the change of blurriness is so small that we can’t even detect it from the image. Hence blurriness is not a strong cue for subtle depth change