The choice of a color camera in a machine vision system depends on the type of application and the objects being inspected by the camera. Color cameras are used for applications where the color of objects is important for distinguishing defects or recognizing details of object orientation for robot guidance, which would otherwise be indistinguishable on a grayscale.

We have developed a vision system for the automatic greasing of hams where it is essential to have a color camera to be able to segment the various components of a ham (meat, fat and rind) which are distinguishable from each other mainly thanks to their color.

In another of our vision systems for Quality control of pizzas with counting of topping ingredients a color camera is used to recognize the different ingredients on a pizza.

In both of the machine vision case studies mentioned above, the image processing software must learn how humans distinguish color nuances and characterize objects by color as well as shape. The goal of automatically processing hams and inspecting pizzas passing at high speed on conveyor belts at a glance, has been achieved through the use of Artificial Intelligence on color images. However, at the heart of any image processing IT solution, whether algorithmic computer vision or neural network training, lies the design of the industrial vision hardware to be deployed.

The sensor of the camera determines the image quality. The most common sensors in color cameras are CMOS (Complementary Metal-Oxide-Semiconductor) sensors, which offer high image resolution and high acquisition speed, as well as low power consumption and low cost.

Color cameras capture images in the RGB (Red, Green, Blue) color space.

The best color cameras contain three independent sensors, each capable of detecting the three color components.

In most commonly used cameras, pixels are made sensitive to a single primary color using optical filters deposited directly on the sensor, using a mosaic arrangement called a Bayer pattern.
In this pattern, filters sensitive to the green component of the color are present in double the quantity of filters sensitive to the red and blue components, to simulate the human eye’s greater sensitivity to the green spectrum.

Multispectral cameras are advanced instruments used to acquire images in different wavelengths of light, outside the normal visible spectrum (which ranges from approximately 400 nm to 700 nm). These cameras use specialized sensors that allow to analyze the properties of materials and objects in greater detail than traditional RGB cameras.

Infrared (NIR) cameras operate in the near-infrared range (about 700 nm to 1400 nm). They can detect imperfections in materials and products that are invisible to visible light.

Thermal cameras (TIR or LWIR): They detect infrared radiation in the thermal range, typically 8 to 14 micrometers (μm), which is related to the temperature of objects. Thermal images display temperatures as a heat map, which can reveal problems that would otherwise go unseen.

UV (ultraviolet) cameras are designed to detect ultraviolet light reflected, emitted, or transmitted by objects and surfaces. Unlike conventional cameras that operate in the visible range, UV cameras require specialized sensors and optics capable of operating at UV wavelengths. These cameras are often equipped with optical filters to select specific wavelengths and enhance the contrast of relevant details.

Automatic X-ray inspection systems are used in industrial processes to inspect parts of products that are not visible from the outside. By analyzing the X-ray image, it is possible to examine the internal structure of an object without destroying it.

In the example image, you can see how a metallic contaminant is easily detected inside a package of dry pasta. There are numerous industrial quality control applications that may require the internal inspection of an object: detecting porosity and inclusions, checking for internal cavities or cracks, and verifying the integrity of assemblies or structural welds.

All colors in the visible spectrum (350-780 nm) can be represented by adding three components of the three primary RGB colors (Red, Green, and Blue). This property is called the additive property of colors. The RGB information set is also called the RGB color space, defined by a cube where the lower-left edge represents black and the upper-right edge represents white.

The RGB space isn’t the only way to represent colors. There are several variations of the three-component color model, designed to better map colors, closer to human perception than the traditional RGB model.

These computational models for image processing are very useful in color machine vision applications for industry

The HSI model separates color into three main components: Hue, Saturation, and Intensity.

HSV model is similar to HSI, but with a main difference in the representation of luminosity. The three components are:

Lab (CIELAB) model is a color space designed to be as close as possible to human color perception and is device-independent. It is based on the theory of human color vision and consists of three main components: