How do you overcome uneven reflections during image processing?

Reflections during image processing

Machine vision is a technology that uses one or more cameras to analyze and inspect objects automatically. Machine vision has gained wide popularity in industrial and production environments. Data acquired by the machine vision system can be utilized in controlling any manufacturing process or activity. A typical application of machine vision on an assembly line is to capture images of parts passing through it, and process the image to detect defects, faults, or anomalies. It can even detect the color, shape, and size of the object for information storage. Apart from defect detection, machine vision is also used in robot guidance, real-time process control, and data collection. However, to overcome uneven reflections during image processing caused due to shiny surfaces, requires careful planning.

Reflections are a common problem in MV image acquisition

Many parts manufactured in the industry have reflective or transparent parts either partially or wholly. These parts also need to be checked for defects or anomalies with equal rigor to maintain the quality of the overall manufacturing process. Transparent and reflective surfaces are often problematic for machine vision acquisition and processing systems. The images captured from these surfaces do not show the clarity and sharpness required for a machine vision system to perform its job adequately. Hence training and inspection of such surfaces pose a huge challenge for the MV model.

What are the reasons for reflections causing challenges?

Uneven distribution of light covering the surface –

A common image-related challenge is the uneven distribution of light falling on the surface of concern. Noise is caused due to external interference and imbalances in the illumination. An uneven illumination causes an incorrect assessment of image attributes around the point where illumination is too much or too little. This in turn affects the performance of the MV model and results in incorrect predictions and missing out on anomalies and defects in the surfaces, compromising on the quality of the product.

Uneven surfaces –

Light falls on the surface or object perpendicular to its plane to achieve even illumination and provide the best possible setting for image acquisition. However, when surfaces are uneven, there will be a few areas that are lit better and a few areas that are inadequately lit. If some fault or anomaly were to be present in such areas, it will be easily missed by the machine vision model. These surfaces will also affect the overall quality of the product due to the inability of the vision system to identify faults. It will reduce the efficiency of the defect identification process.

Working distance –

The working distance of any machine vision system depends on the object getting captured and the mechanics surrounding the object getting captured. An assembly line works through the harmony between the human operators and the robots that employ machine vision tools for image processing. The position of cameras relative to the object and one another plays a very important role in outputting an image of superior quality. There is a minimum working distance constraint and maximum working distance constraint and these affect MV processing.

How are such challenges overcome?

Use of uniformly distributed light –

Uneven light distribution causes concentrated pockets of light where illumination is too high and on the opposite side, illumination is too low. Reflections created due to such a distribution of light interfere with the machine vision model. More or less light falling into the camera can change the image drastically. Using a uniformly illuminated lighting structure like a dome light can tackle this issue. A dome light is a hemispherical lighting tool with LED lights on the circumference. The body of the dome is a white reflective surface that indirectly lights up an object instead of creating a single source of concentrated light that gets scattered over the surface of the object at hand.

Use of polarizers –

As the name suggests, a polarizer is used to polarize light. The function of a polarizer is to filter different light sources and admit light rays from different sources only if the light rays are coming uniformly from one direction. There are two kinds of polarizers, linear and circular. Depending on the needs of the project a single one is chosen for absorbing non-uniform light rays linearly or radially. Polarizers are ideal for cases where there is excessive light falling from various directions onto the object.

Use of AI –

The issues of reflections and specularity are preferably dealt with at the light source itself. The use of dome light and polarizers can solve the issue and produce a well-aquicized image. If there is a scenario where both these tools are not usable, then using AI and deep learning models can also be involved in resolving this issue. When there is a training subset consisting of all reflective surfaces, the machine vision can learn to treat such images just the same as the ones getting clicked in the uniform lighting conditions. A combination of varying degrees of light source illumination and camera work can add to the variety of conditions that a machine vision system will encounter in a real-world setting. This will fix any remaining issues that a reflective or transparent surface with uneven lighting conditions might create in the image processing.

Thus, uneven illumination, reflective and transparent surfaces are present in many parts in the manufacturing domain which pose a real problem to machine vision systems. These surfaces result in compromised image quality and prevent the algorithms from finding faults in the affected areas. To solve this issue, the usage of uniformly distributed light and polarisers is done to prevent any issues from the point of image acquisition. If the image issues persist, machine vision systems can also be trained to identify such reflections by extensive training and learn the ability to find defects and make predictions from the compromised images as well.

Share and spread the knowledge

Leave a Reply

Schedule A Demo
close slider