IEEE Future Directions
Related Articles
I still remember my first film camera (I was really proud of having a camera!). It had an optical indicator that as I turned the focus ring on the lens showed a detailed area in the frame split in two that recomposed into a single image once the focus was achieved. I should say that is was pretty accurate, the only drawback was the time it took to rotate (slowly) the ring to the point that focus was achieved.
Current digital cameras use a very similar approach to focus: phase detection. Samsung has just perfected phase detection by working at the camera sensor level. Each pixel is split into two parts: the light received by each is processed by a circuit that looks at phase differences and drive the focus ring of the lens to a position where the incoming beams (in a specific area) are overlapping, meaning the are reflected from an object that is now perfectly in focus. This mechanism was developed by Canon. What Samsung announced is a perfected way of detecting phase by splitting some pixels diagonally, and others vertically (Canon splits pixels only vertically). The problem with splitting pixels only vertically is that if you are taking a shot at a surface that has only horizontal lines that vertically split pixels are not able to detect any phase displacement (i.e. they get confused and cannot focus). On the other hand, if you have both vertically and diagonally split pixels you can rest assured that either one of the other type will detect phase difference and hence will be able to focus (I had the same problem with my old camera and what I did in those -few- cases where focus was tricky was to rotate the camera to get the focus and than rotating it back to take the photo).
You can get a clear explanation of the way the new Samsung Dual Pixel Pro sensor focus works by watching the clip..
Another interesting feature of this new chip is that the 50 Mpixels of the sensor can be clustered in groups of four (thus decreasing the overall resolution to 12.5 Mpixels) to harvest more light, something that comes handy in situations, like night photography, where there is very little light and the noise becomes an issue. There is more! Since each pixel is actually split in two parts the sensors also support the transfer of data from each half part, thus providing an equivalent of 100Mpixel sensor (but, of course, the signal to noise ratio gets worse and it makes sense only when there is plenty of light). Interestingly both the clustering of four pixels into one as well as the splitting of one into two are accomplished by the sensor circuit so that it is not affecting the performance of the camera electronics. This is really important since the sensor is designed for smartphone cameras where the size of the sensor is constrained by the size (thickness) of the smartphone.
Really impressive.
