Basic principles of image sensors

An increased resolution makes it possible to see better the details of the scene, but sometimes to the detriment of the transfer speed. A camera set at \(200\times 100\text{ pixels}\), that is to say at \(20,000\text{ pixels}\), has to send \(20,000\) numerical values to the acquisition system. If the system works at \(25\text{ MHz}\), it corresponds to \(40\text{ ns}\) per value, that is to say \(0,0008\text{ s}\) for a full image, which equals \(1250\text{ fr/s}\). If switching to \(640\times 480\text{ pixels}\), that is about \(15\) times as much pixels, the acquisition frequency has to be reduced to about \(80\text{ fr/s}\).

According to the displacement speed of the object, appropriate exposure times and acquisition speeds should be given priority. The exposure time makes it possible to fix the moving object, for example a fast-moving small particle. As for the acquisition speed, it makes it possible to get several fixed images of the particle in the same field of vision. This is particularly necessary for correlation or tracking. This kind of capture can also be done using appropriate lighting (pulsed laser, stroboscope) synchronised to the camera. At a last resort, the blurred length of the (very rapid) particle path during the exposure time makes it possible to determine its speed in only one image.

The frame rate of a camera determines its ability to record a series of full images in a given time. It is typically expressed in frame per second: \(\text{fr/s}\). This frame rate depends not only on the matrix size, but also on its architecture. Moreover the frame rate depends on the type of connection between the camera and the signal acquisition system. The fact that it has or not some on-board memory also influences the frame rate. For example, if the format is the same, it is possible to transfer \(640\times 480\text{ pixels}\) at \(50\text{ fr/s}\) (CCIR mode) or at \(200\text{ fr/s}\) (IEEE1394b), or even several \(text{kfr/s}\) in on-board memory or in SATA link.

The spectral response mainly depends on the type of semiconductor and on the sensor treatment (sensor lighted from the front or from the back, slimmer sensor or not). Typically, all the sensors have a response between \(400\text{ nm}\) and a bit less than \(1000\text{ nm}\). Nevertheless, it has to be borne in mind that a camera has a given response at a given gain, and that a high response obtained at a high gain may generate a very high noise, reduce the dynamic range and thus damage the image quality.

During the digitization of the charge signal, the full-well capacity is discretised in greyscale. This well depth can be encoded on \(8\) or \(16\) bits, or even more. The \(0\) level corresponds to \(100\%\) of black and the \(255\) level to \(100\%\) of white, and similarly for \(10\) bits (\(1,024\) levels) and \(32\) bits (\(4,096\) levels). Of course, the weight of the pixel influences the quantity of information to transfer (in Mbit/s), and this has to be taken into account for the bandwidth of the transfer system.