Deep learning for automated image analysis

Image segmentation is used to analyze images taken with a microscope. Segmentation refers to the division of images into specific areas that are important for subsequent analysis and classification. Such an area could be, for example, a defect or contamination on the surface of a component, as well as the detection of different material layers. During the subsequent analysis of the images and classification of the recognized areas, the areas themselves and the boundary between different areas are considered. This enables accurate results to be delivered and errors to be detected.

However, traditional methods of segmentation, such as thresholding (gray value analysis), quickly reach their limits.

The gray levels of the areas can be difficult to distinguish if they have a similar color and brightness. Users are also faced with the question of which features in the image are relevant, e.g. color, texture or edges, in order to identify objects and areas in an image.

It is also important to know how to combine the features in order to discover objects and classes. The more classes are added when processing an image, the more complex the task becomes. The search for scratches on displays for electrical appliances is also a challenge that is difficult to solve with rule-based analyses - because every scratch is different in size, has its own shape and can occur on the entire surface. Image processing with deep learning is the right solution here.