This topic defines digital camera parameters, as they relate to image quality, and explains how they are measured at Pixelink. There are many more but we believe, in our humble opinion, that the following parameters (listed in alphabetical order) are the most important.


Dynamic Range


Dynamic range is a measure of the ability of a camera to reproduce contrast within a scene. Pixelink publishes electrical dynamic range, which is defined as follows:



where MaxOutput is the maximum output available from the camera and ReadNoise is the read noise. On occasion,dynamic range is quoted as a ratio instead of in decibels.


Example


Using a 10bit camera (MaxOutput =1023DN), the EDR will be 54dB if the read noise is 2DN. 

In reality, you need to consider the impact of the FPN and swing possible with the given lighting. FFC can help improve the dynamic range by effective MaxOutput.  See TN 004: Interpreting Signal-to-Noise Measurements on dynamic range for a discussion in greater detail.


Fixed Pattern Noise (FPN)

FPN is a fixed pixel-to-pixel offset and is due to a combination of variations in the image sensor amplifiers and dark current. It is fixed with respect to the illumination; however, it is temperature, integration time, and gain dependent. FPN also depends on the video format (e.g. it will be less for 10bit Bayer data than for Y’CBCR data). For CMOS based cameras, there may also be a brightness dependency. It is measured at a given integration time, temperature, and gain under dark conditions.  For megapixel area arrays, FPN is usually quantified as follows:



where sigma is one standard deviation of the camera variation, and MaxOutput is the maximum output available from the camera.


Example


sigma=5.2DN, MaxOutput =1023 (for 10bit mode). Then the FPN is on the order of 0.5%. As with PRNU, FFC can be used to reduce FPN. 


Photo Response Non-Uniformity (PRNU)


PRNU is a pixel-to-pixel variation in the sensor responsivity. Consider it analogous to FPN: FPN is an offset; PRNU is a gain. It arises due to variations in pixels. For example, although a sensor may be quoted as having a 3.5mm by 3.5mm pixel, in reality, there will be a variation in pixel size across the array. It is measured at a given integration time, temperature, and gain under uniform lighting conditions (typically at half of the saturation level - 512DN in 10bit mode). This area variation will contribute to PRNU. For megapixel area arrays, PRNU is quantified as:


 

where sigma is one standard deviation of the camera variation, and FrameMean is the mean value of the frame of video. Some vendors normalize by MaxOutput instead of FrameMean, so care needs to be exercised when comparing camera specifications. As with FPN, FFC can be used to reduce FPN. Pixelink takes great care to assure the uniformity of the light used to measure PRNU. Non-uniformities in the light will contribute to a phantom PRNU. PRNU also depends on the video format.


Noise Equivalent Exposure (NEE)

NEE is the ratio of the read noise to responsivity. NEE is a measure of the exposure to produce a signal-to-noise ratio of one. The smaller the NEE, the better the camera. Pixelink quotes the responsivity at 600nm.  Some vendors use average or peak responsivity. Again, when comparing NEE, one needs to be aware of what type of responsivity is being quoted.



Read Noise

Read Noise is the uncertainty in the value of a pixel. If a pixel is repeatedly sampled, then there will be an intrinsic uncertainty in its value. Read noise quantifies this uncertainty. There is no commonly adopted way of measuring read noise in the industry. Pixelink measures read noise by sampling the center 64 × 64 array of pixels 32 times. The rms noise of each pixel is calculated (forming essentially another 64 × 64 array of pixel standard deviations). The average standard deviation is then calculated. Read Noise will scale with the camera gain.


Responsivity

In optics, responsivity is defined as the output (in digital numbers, DN, for a digital camera) divided by an input (areal energy density - in nJ/cm2). This is a spectral varying parameter and accounts for how the sensor responds to light of varying color. It varies with wavelength (or the color of light) because the absorption coefficient varies with wavelength. Blue light gets absorbed near the top of the silicon photoelement, while red light gets absorbed deeper in the photoelement.


For a color camera, there are three sets of responsivity curves (one for the red, one for the green, and one for the blue channels). The responsivity curves shown are for unity channel gains. Doing a white balance will scale each of the responsivity curves by the channel gain.


The responsivity will scale with the gain.


Saturation Equivalent Exposure (SEE)

SEE is the ratio of the maximum available signal from a camera to the responsivity. It is similar to the NEE. The larger the SEE, the better the camera. Pixelink quotes the responsivity at 600nm. Some vendors use average or peak responsivity. Again, when comparing NEE, one needs to be aware of what type of responsivity is being quoted.