29865 Six Mile Road
Livonia, Michigan 48152 USA
Products and Solutions for Process Improvement
Machine Vision Components, Systems and Turnkey Inspection Solutions
Sensor sizes range from ¼” to the new 1-1/8” of the 22 or 29 Megapixel sensors, the Pixel (Picture Element) is the active site which is used to accumulate photons. When even greater resolution is required for an application the implementation of line scan sensors is employed. Phoenix Imaging has implemented line scan sensor up to 16K is size producing extremely high resolution images. The more pixels a sensor has the great the resolving power of the image or put another way, the more pixels per image area the higher the resolution. A simple way to calculate the resolution of a sensor is to divide the sensor size by the number of active pixels. The illustration below show the relative size of the various sensors available and their physical dimensions. It should be noted that the sensor size is actually smaller than it's name implies, i.e. 1/2" sensor is not 1/2" in any dimension. Sensors (CCD / CMOS) are often make reference to an imperial fraction designation such as 1/1.8" or 2/3", this measurement actually originates back in the 1950's and the time of Vidicon tubes. Those who find the specification sheets for these sensors are then even more confused about the relationship between the fraction and the actual diagonal size of the sensor. These sizes were typically 1/2", 2/3" etc. The size designation does not define the diagonal of the sensor area but rather the outer diameter of the long glass envelope of the tube. Engineers soon discovered that for various reasons the usable area of this imaging plane was approximately two thirds of the designated size.
The sensors are typically manufactured as CCD (charged coupled device) or as CMOS (Complementary Metal Oxide Semiconductor). The CCD has been around the longer than the CMOS and offers a few advantages over the CMOS technology, such as the CCD sensors are more sensitive than CMOS resulting in better images, especially in low light conditions. The CCD sensors are cleaner or less grainy than CMOS sensors because they less susceptible to noise (electronic). The CMOS sensor are less expensive to manufacture because they are based on the same technology used to manufacture most electronic devices, i.e. computer chips. Since the CMOS technology is similar to most electronic devices the manufactures can actually integrate electronic circuitry on the CMOS sensor element to improve the signal. This integration of electronics in the sensor area (pixel area) requires real estate that would have been used for gathering photons, thus reducing the "active pixel size". Most consumer cameras today use the CMOS technology and great improvements are being implemented to produce images that are close to the quality of CCD sensors.
The old Vidicon tubes were circular in shape and the image used by the tube was taken from the center of the tube. Today most sensors do not have a square shape, rather they are rectangular in shape. Early CCD and CMOS sensors had a 4:3 aspect ratio where the horizontal axis was 4 units long and the vertical axis was 3 units high. Today you can find sensors in a number of shapes from square (1:1) to Widescreen (16.9). The longest dimension of the image produced by the optical system (lens) must fit within the image circle of the sensor. The image produced by the optical system (lens) produces an image of the object at the focal plane of the optical system.
Assuming that you can get the object to fit on to your sensor correctly we also have to consider the dead space on the sensor. This is important if you do not want objects of interest to literally "fall in the cracks" of the sensor. All sensors have space that separate the pixels. This shape appears as lines on sensor and produces a grid pattern of pixels. This space or line width between the pixels is very similar for all sensor types and must be considered when selecting a sensor size for your application. Since this line width is about the same it only goes to show that smaller sensors use a greater percentage of the sensor area with this dead space than is found on larger sensors. If the defective condition that we are attempting to isolate is small it has a greater chance of "falling into the crack" of a small sensor than that of a large sensor. This is often observed when a sensor or a component under test is moved slightly and the defect "disappears". If the component or sensor is moved again a small displacement the defect "reappears" in the image area. This problem exist for both large and small sensors but is more prevalent in the smaller sensors.
You can think of each pixel as photon collector or a photon bucket. The more photons that a pixel collects the greater the charge on the element and brighter the pixel appears in the image. The sensor is designed with a maximum grayscale sensitivity, typically sensors have a minimum grayscale resolution of 8 bits (0-255 gray levels). Better sensors often provide 10, 12 or 14 gray levels. A 10 bit system offers 0-1023 grays levels for image depth. This means that the image of 10-bit system is 4 times more steps than that of an 8 bit system. This is important because the background noise of a 50 dB signal to noise ratio sensor is about 5 to 10 grays levels. That means every time that you acquire an image using this sensor the gray level of an individual pixel can change 5 to 10 gray levels or approximately 2% of the total range of the sensor. If you use a vision algorithm with a fixed threshold and the gray level of defective condition is near your threshold it may be isolated as a defect in one image acquisition and then as acceptable in the next image acquisition. If you use a sensor with 10-bits of grayscale resolution (depth) then the same 5 - 10 gray level variation represents only about 0.5% of the full scale range. This disadvantage of the using a sensor with greater grayscale resolution is that the image is now requires 16 bits of data per pixel rather than the 8 bits. This leads to greater processing time for an equivalent sensor area (horizontal x vertical pixels). However, it is often more efficient use of image processing cycle to start with a better quality image than to perform more operations to "clean up" a noisy image. The illustration to the right shows the relative contrast of the various level of grayscale, 0 gray level equals black and 1023 gray levels equals white (saturated pixel). It should be noted that the typical human eye has the capability to differentiate about 40 - 45 gray levels when show multiple gray level targets (targets with a range appear to be the same gray level).
Another important factor in a system ability to perform surface inspection is "Background Normalization". This function can not be performed with the aid of image processing, i.e. the human inspector does not have the capacity to perform this operation without the use of a computer or image processing system. The simplest form of this function is represented by the subtraction of a background image from an image that contains the same background information plus some additional information. The resultant of this operation is only the information that was not present in the background image. The illustration to the left represents the three images, Sample Image + Background Image, Background Image and the resultant Sample Image. Granted the application is more complicated than this simple example but the principle is the same except that the comparisons are performed on pixel neighborhoods.
This is very simplified example of the function will only work for basis images in which low contrast information is not important. A more sophisticated approach is to create the "Background Image" form the actual sample image using appropriate image processing functions. This is part of the unique image processing functionality of the Phoenix Imaging AVSIS™ surface inspection algorithms. We compare each pixel in the sample image to the surrounding pixels so that defective regions are isolated based on relative contrast to the surrounding areas. This allows the system to isolate defective regions even if the lighting conditions change across the component under inspection.
The isolation of small imperfections that have very low contrast must be performed using high bit depth sensors. Sometimes the imperfections are only 20 or 30 gray levels above the background on a 4096 deep grayscale image. If one is using an 8 bit sensor or image processing system the defect may not even be visible. More bit depth results in larger images because the image is represented by a 16 bit word. 16 bit processing usually requires more time than the smaller 8 bit deep images therefore the amount memory and processor speed become critical.