Machine vision has become the eye of industrial automation. With a machine vision system, repetitive and arduous tasks can be accomplished with tremendous accuracy and high speeds while drastically reducing costs. No wonder it is rapidly gaining traction in many industries. But how does the machine vision system actually accomplish these tasks?
Well, a machine vision workflow can be broken down into the following four major processes:
- Pre-process automation
- Image acquisition
- Image processing
- Post-process automation
In this post, we will majorly explore the image processing part of the machine vision workflow and address some important questions related to images.
What is Image Processing?
Image processing is the process of applying software algorithms or some operations on an image to achieve an enhanced image or get the desired results. Basically, it is a kind of signal processing step that takes the input as an image, analyzes or manipulates the image and outputs an altered image or some extracted data associated with the image. After the image has been acquired, two components are necessary for the image processing part, which are: image processing software and computational hardware
Related Article– IMAGE PROCESSING
Vision systems deal with algorithms for automating visual perception and might involve noise removal, sharpening of edges, segmentation, object detection, interpretation of the scene, to name a few.
How Are Images Formed?
Images can fundamentally be understood as a matrix of miniature potential wells called pixels. These pixels can store particular values that correspond to the associated intensity. For example, let us analyze a black and white image. The pixels in a monochrome store a value from 0 to 255. 0 corresponds to black color or an opaque pixel. Similarly, 255 corresponds to maximum transparency or white color.
Let us now consider color images. We know that when the three primary colors, namely red, blue and green, are mixed in different proportions, they form all the different kinds of colors. Each of the pixels in a color image can be represented as a scalar number or vector with three values corresponding to each of the three primary color’s intensity. These RGB values are responsible for the characteristic color of the pixel and, in turn, the entire image.
Relation Between Image Size and Resolution
Firstly, what is resolution? Resolution can be understood as the amount of object detailing or clarity of your features. To find out the resolution requirements for your machine vision application, you need the following three parameters:
- Minimum feature size (MFS)
- Pixel resolution (PR)
- Field of view (FOV)
The formula to calculate resolution along any one dimension is: (FOV X PR)/MFS.
Let’s say your camera has a specification of 4MP. So, what is the number of pixels across each dimension? Well, the answer depends on the resolution across each dimension. It can be 2000px & 2000px or 4000px & 1000px, or any combination where the multiplication of resolutions offered along each dimension is 4 MP.
Also, Read PARAMETERS FOR LENS SELECTION
Another interesting question is: How can you estimate the typical image size of a monochrome image captured by a 4MP industrial camera? As a rule of thumb, one byte corresponds to the data in one pixel. An image captured by a 4MP camera will have 4 x 106 pixels and, thus, will commonly have a size of 4MB. But will it still be a 4 MB image for a color camera? Intuition suggests that a color image stores more information than a black and white image. This is a cent percent true. As we talked earlier, pixels in color images have store three values while pixels in monochrome images store only one value. Therefore, since color images store three times the information, its size will be 12 MB.
Image processing is a crucial stage in the machine vision workflow. At this stage, the software implements certain algorithms to provide you with the desired pieces of information or alter the image according to your needs. Getting this part right is essential to the vision system’s accuracy.
In this blog post, we discovered the basics of image processing and analyzed how an image is constructed. We also understood how exactly the image resolution and the image size are correlated.