Autofocus Systems Part III

  • Contrast Detection Autofocus

With the widespread introduction of digital cameras in the late 1990′s, a new autofocus technology was developed. Contrast detection autofocus performs a continuous reading of the scene during focusing. This technology is based on a constant evaluation of data provided by the image sensor before the actual picture is taken. This implies that no camera part may be located in front of the image sensor while autofocus is performed. For that reason, contrast detection autofocus is primarily used in mirrorless digital point-and-shoot cameras. Also, this autofocus concept is of major importance for notebooks, smartphones and tablets due to serious space limitations. On DSLR cameras, contrast detection autofocus is only available in live-view mode where the reflex mirror is locked up.

The image sensor serves as a contrast detector that successively detects contrast data of a subject to be photographed. A focus detector comprises electronic circuitry that is used to calculate a difference value between currently detected contrast data and previously detected contrast data, and finds a focused situation accordingly.

Perfect focus usually coincides with maximum contrast on the image sensor and therefore, this type of autofocus measures the intensity of contrast data within a selected focusing region. The following illustration shows the relation between focus and contrast.

Contrast Detection

Here is a step-by-step description of this iterative procedure for the determination of focus with the contrast detection technology: In a first step, the focus detector circuitry receives a snapshot of the contrast situation provided by the image sensor. However, there is the problem that no information is available about whether the current snapshot already represents the strongest contrast or if the current focus situation can be optimized. Also, if low contrast is detected right at the first snapshot, the system doesn’t allow conclusions on whether the low contrast results from the focal point being in front or behind the image sensor and therefore, no direction can be identified at this step.

For these reasons, the focused position is usually detected by a gradient method – called the »hill climbing method« – in which a peak in a curved line of contrast values is searched. In the second step, the focusing lens is driven in a predefined direction while successively taking snapshots of the contrast situation in order to determine the orientation of the gradient. If the contrast decreases, the system changes the direction of the lens movement immediately. As long as the contrast increases, the lens keeps moving until the contrast data has a peak value. To confirm whether the contrast data is actually at its peak, the focusing lens is driven so as to surpass the focused position, detecting lower contrast again. Consequently as a last step, the focusing lens is driven back to the position that has produced the peak signal. The following diagram shows the hill climbing method.

Hill Climbing

Focusing Region Selection

It is necessary to limit the active focusing region so that not the entire image sensor is evaluated but only pre-selected focusing zones. A reason for the use of smaller focusing zones is that a scene very often includes areas that deliberately should remain unfocused, such as undesired background or other elements that might distract the viewer from the actual subject. If the contrast detection autofocus system reads the entire image sensor for focus detection, any large-scale areas would be prioritized, leaving the actual subject unfocused.

Compact digical cameras usually deploy smaller focusing zones in the center of the image sensor. These zones are typically displayed on the LCD display of the camera as square indicators. Depending on the type of autofocus system, most compact digital cameras allow to change the positions of these focusing zones, or allow the camera electronics to change it automatically. For instance, face detection systems perform a scan of the entire scene and prioritize regions where faces have been recognized.

Conclusion

Until the last couple of years, contrast detection autofocus has been the only option for mirrorless cameras to achieve sharp images. It is considered to the the most accurate type of autofocus as it evaluates the signal from the light-receiving surface of the image sensor directly. Furthermore, no additional autofocus sensor unit is required, which reduces production cost. Finally, contrast detection autofocus systems are not susceptible to calibration errors such as front or back focus which can occur with phase detection autofocus systems.

On the other side, contrast detection autofocus does not involve actual distance measurement at all and therefore no directional information is available for the autofocus system. This makes contrast detection autofocus slower and speed decreases even more in dim environments with low contrast. Also, the constant illumination of the image sensor increases its temperature and can therefore increase noise. It has to be noted, however, that contrast detection autofocus systems have improved significantly over the past years.

  • Hybrid CMOS AF

The next step in the evolution of digital camera autofocus was a technology where contrast detection autofocus was supplemented with elements from phase detection systems. Canon calls this technology Hybrid CMOS AF. With this new autofocus system, a new sensor technology is used where the active light-receiving surface includes a certain number of photodiodes that are specialized to perform phase detection.

The specialized photodiodes are designed so that they only receive light from one half of the lens. This is achieved by applying a tiny cover plate over one half of the light-sensitive depletion region. As a result, light coming from one half of the lens is effectively blocked from reaching the photodiode. Thus, depending on the position of the light cover plate, a photodiode can either receive light from the left half of the lens or from the right half. The following image shows this special design of a photodiode.

Hybrid CMOS AF Photodiode

Still, such a specialized photodiode cannot perform phase detection individually. It is required to have a relatively large number of these cells integrated into the image sensor with an equal distribution of left-looking and right-looking pixels. Then, the autofocus system compares whether the image produced by all left-looking pixels is in the same position as the image produced by the right-looking pixels. If there is a shift between these images, the system determines both the direction in which the half-images are shifted and the distance by which these are separated. This information is what allows Hybrid CMOS AF to drive the lens to the in-focus position in a very targeted way.

Hybrid CMOS AF Phase Detection

However, any phase-shift of both half-images does not leave these images unaffected. The larger the phase shift, the further the focal point is from the sensor, thus producing blurrier images. To ensure accurate predictions nonetheless, the layout of the image sensor must be capable of detecting strong deviations. There are various implementations of Hybrid CMOS AF ranging from uniform distributions of phase detection pixels up to straight lines of phase detection cells. The figure below shows a layout where phase detection pixels are placed uniformly distributed on the surface of the sensor.

Hybrid CMOS AF Sensor Layout

Although a fair number of phase detection cells is evenly distributed over the image sensor, their resolution is still not very high. For that reason the system is unable to detect slight focus deviations when the in-focus position has almost been reached. Rather, the additional information acquired by the phase detecion component is used to indicate just roughly where the maximum contrast is located so that the lens can move in the right direction from the beginning. For fine adjustment around the in-focus position, contrast detection is performed with the hill climbing algorithm. As a result, Hybrid CMOS AF is a combination of high speed and high precision focus. Not being as fast as a full phase detection system, this is still a huge step forward in the development of autofocus systems for digital cameras. The following diagram shows how targeted the lens can be moved at the beginning, and how hill climbing is performed towards the end.

Hybrid CMOS AF Hill Climbing

After the introduction of Hybrid CMOS AF in 2012, initial improvements were soon applied, and there already are three generations of Hybrid CMOS AF for Canon DSLR cameras. On the Canon EOS 650D (Japan: EOS Kiss X6 / North America: EOS Rebel T4i), released in 2012, Hybrid CMOS AF (I) was used. This generation had a small portion (10%) in the center of the image sensor covered with phase detection cells. In the following year, the Canon EOS 100D (Japan: EOS Kiss X7 / North America: EOS Rebel SL1) was released with the Hybrid CMOS AF (II) installed. This second generation already had 60% of the image sensor covered with phase detection pixels and had an improved performance for autofocusing while shooting video. Finally, the Canon EOS 750D (Japan: EOS Kiss X8i / North America: EOS Rebel T6i) was released in 2015 that had Hybrid CMOS AF (III) applied with even further improved video focusing performance, a higher density of autofocus pixels that were distributed in a regular pattern and which was four times faster than the second generation, according to Canon. The image below summarizes the different generations of Canon’s Hybrid CMOS AF.

Canon Hybrid CMOS AF Generations

The disadvantages of Hybrid CMOS AF systems are, however, that they reduce the sensitivity of the specialized pixels due to the cover plates. Depending on the camera electronics, these pixels can either be used for phase-detection only or contribute to the final image. If they are not read out for the final image formation, the respective signal of this pixel has to be generated by interpolation. Also, the readout circuitry needs to be specialized to allow separate readout cycles for phase detection and for the remaining pixels. Both the enhanced readout logics and the implementation of specialized phase detection cells increase the production cost for these systems. As they are still not able to compete with full phase detection systems, today’s DSLR cameras still deploy dedicated phase detection sensor units underneath the mirror, which is why Hybrid CMOS AF is more an improvement for video shooters and live-view photography.

Continue to Part IV