Texture inspection is the analysis of the surfaces such fabric, leather, rubber, wood and so on and the process of finding defects. Along with finding a common use area in the textile industry, there are applications in many different sectors from the metal industry to the plastic industry, from the food industry to the packaging industry.

Texture inspection, also known as novelty detection in the literature, is one of the most difficult areas of image processing. It can often be difficult to tell the computer by modeling mathematically the defects that can easily be recognized by the eye. Since defects are not always in the same form, a general approach will not always work.

Even though it is very easy, in some cases, to find mistakes on a fabric, wood, leather, or rubber surface, a program that can find almost any kind of defects has to use advanced artificial vision techniques.

In places where there is a single type of production, certain types of errors are easy to find, but usually the manufacturers are generally working with models and patterns in very different types, due to their dynamic production models. It makes it almost impossible to develop a general algorithm since the surface to be controlled (fabric, wood, leather, etc.) and the defects to be recognized (oil, stain, color difference, tear, cut, etc.) are both variable.

For example, in the above picture, the faulty regions can easily be seen visually. With the classical approach, it is easy to find black areas and perform field control as simple as possible, or catch errors by looking at the diameter feature. However, if different colors and designs are to be produced, the field and color controls will have to change each time, and in some cases these controls may label a faulty product as not faulty or vice versa.

As AKU, we use 'Classification Based Machine Learning' approach for such cases. In this method, 4-5 different filters are applied on the picture (Generally taking 1st and 2nd derivatives). The obtained softened and Gaussian filtered pictures are subjected to classification process. Then, the system will automatically detect each type of fault (without having to enter any parameters like color, height, area).

Generally, classification methodologies known as GMM (Gaussian Mixture Model), MLP (Multi Layer Perception), SVM (Support Vector Machine) are applied. Recently, along with the development of deep learning technologies, CNN (Convolutional Neural Networks) based approaches have been also tried, but they are less preferred since they need much more sampling.

As you can see, in the texture inspection problems that seem to be easy at first but can become quite complicated when it gets into the work, we use the simplest but technologically advanced methods as far as the end user is concerned.

The images can be illuminated from the front or back according to the type of the fabric and the desired catch, and line scan or area scan cameras can be used.

The systems we build by doing texture inspection are usually quality control systems that catch the following defects.

• Weft-warp Faults
• Oil, dirt, etc. stains
• Pattern Disorders
• Color Differences
• Faults like tear, melting
• Faults like insect bites, surface wounds, scratches
• Faults like swelling, collapse

For these defects, precision is determined according to customer needs. Camera, lens and lighting are selected at a resolution suitable for customer sensitivities. Often multi-camera systems are being installed. Production speed is also an important parameter to consider.

As a result, a solution based on the type of error required to be caught is presented. The solution we offer is a machine vision solution that can learn by itself, that can adopt to a variety of different patterns, and that uses deep learning techniques as well.

Surface inspection is similar to texture inspection methods. In general, the surface is not as complex as the texture. Widely used in white goods, glass, automotive and metal industries.

Main areas of use is to detect faults such as

• Surface scratches
• Cracks
• Fractures
• Swelling or collapse

Finding defects on the outside surfaces of white goods such as refrigerators or washing machines, controlling the paint on surfaces such as automobile body and bumpers, and looking for broken, torn, pierced areas on metal sheets exposed from the press are some examples of surface inspection usage.

If the surface is planary (as in the case of a refrigerator) 2-D analysis methods are used. if the surface is 3D (usually metal sheet surfaces) surface inspection is difficult. In this case the most appropriate solution is sought. By means of alternates such as:

• 3D Camera
• Multi Camera Setup
• Moving the camera over the robot gripper

 

Metal parts with surface scratches (left) and caught faults (right)

In surface inspection, filtering techniques in image processing are used. Transformations such as Fourier transform, inverse fourier transform, convolution, and gaussian filters are commonly used.

Lighting will play a greater role than ever in finding surface defects. Especially Photometric Stereo and Deflectometry methods make it possible to catch many mistakes that are difficult to see even with eyes.