Image processing
In computer science and electrical engineering, image processing is the processing of signals that represent images, such as photographs or individual images from videos. The result of an image processing can in turn be an image or a set of features of the input image (see image recognition ). In most cases, images are viewed as a two-dimensional signal, so that common methods from signal processing can be used.
Image processing is to be distinguished from image processing , which deals with the manipulation of images for subsequent display.
Application examples of image processing
Image processing is now used in almost all scientific and engineering disciplines, for example in modern microscopy , medical diagnostics, astronomy , mechanical engineering and in remote sensing (environmental observation, espionage). Image processing methods are used to count and measure objects in machines, inspect objects or read coded information. X-ray and ultrasound devices use image processing to provide images that the doctor can interpret more easily. X-ray machines with integrated image analysis automatically examine luggage and clothing for dangerous materials and objects (weapons, etc.) in security zones. Another field is quality assurance in manufacturing and production processes . The so-called picking in the box in robotics is also supported by image processing.
Inspection and measurement of objects
Measurement of chips in semiconductor production : For example, the position of the corner of a chip is measured in an image. With this information, this chip can be precisely positioned for assembly. During the final chip inspection, soldering and bonding errors are detected by moving a chip in front of the camera and comparing it with a golden sample that was previously taught in by the image processing system. Or it is searched for clearly defined geometric patterns (circle, corner), for cracks or breakouts.
Image processing in a beverage filling system: To check whether the same amount has been filled in each bottle, an image of the bottle neck is taken and the liquid edge is measured. Before filling, a check is made to see whether the neck of the bottle has cracks or splintering.
Measurement of glue in micromechanics: When manufacturing camera modules for cell phones, glue is applied to the lens holder. In order to ensure a consistent quality of production, the shape of this adhesive is checked in a picture. If the adhesive application is not within certain tolerances, the component is rejected.
Reading coded information
Using image processing, coded information can be read out automatically from images. For example, a text coded in DataMatrix format can be read out or information can be extracted as plain text using OCR . These functions are also used in letter and parcel identification.
In automotive production, serial numbers of components are encoded in data matrix form. When an assembly reaches a production area, a camera takes a picture of the code and the code is read out. With this serial number, machines in the production area receive information from a server on how the assembly is to be handled.
Image processing objects
Image processing methods generally expect image data as input. These image data can be differentiated both in the way they were created and in their coding . The type of creation describes the technical principle used to create the image. The most widespread here are reflection images such as those produced by a camera or ultrasound. In addition, there are projection images , such as X-ray images , and schematic images , such as maps and documents. The most common form of coding is raster graphics , in which the image data is represented by a two-dimensional grid of pixels. Another form is vector graphics , which do not consist of a grid but contain instructions on how to create an image from geometric primitives.
Delimitation from related areas
Related areas of image processing are image processing , computer vision and computer graphics . With image processing , a more abstract view of the change in images is laid, while image processing provides the mathematical and algorithmic basis for this, which is then used in the implementation of graphics software for image processing. Image processing also provides this for computer-aided vision . While image processing generates image data or simple information from image data, computer-based vision generates image descriptions from image data. The computer graphics in turn generate image data from image descriptions.
Image processing operations
In image processing operations, a distinction must first be made between processes that generate a new image and those that provide information about the image. The methods that generate a new image can be distinguished based on the size of the region of the input data. A distinction must also be made as to whether the process preserves the basic structure of the image or changes it.
A common method of generating information from an image is to calculate the histogram , which provides information about the statistical distribution of brightness in the image. Such a histogram can, for example, serve as a configuration for further image processing steps or as information for a human software user. Further calculable information about an image is, for example, its entropy or mean brightness.
Methods that generate a new image can be divided into point operations , neighborhood operations and global operations on the basis of their input data . The point operations use the color or lightness information at a given point in the image as input, calculate a new lightness value as the result, and store it at the same point in the target image. Typical applications of point operations are, for example, the correction of contrast and brightness, a color correction by rotating the color space or the use of different threshold value methods . A point operation can either be homogeneous , which means that the coordinate of the source data is not taken into account in the calculation, or they can be inhomogeneous , which enables adaptive tonal value correction, for example. Neighborhood operations use both a point and a certain set of its neighbors as input, calculate a result from them and write this to the coordinate of the reference point in the target image. Convolution filters are a very common type of neighborhood operation . Here, the brightness or color values are offset against one another according to a filter core in order to form the result. With this method, for example, soft focus filters such as the mean value filter , Gaussian filter or the binomial filter can be implemented. Convolution filters can also be used to highlight the edges of an image using derivative filters or Laplace filters . However, the neighborhood operations are not limited to the convolution filters. With more complex algorithmic treatment of the reference point and its neighbors, further smoothing methods such as the median filter or the extreme span filter or the Prewitt operator for edge detection can be implemented. The operations opening , closing and thus morphological smoothing can be defined from morphological operators such as erosion and dilation . While smoothing can be achieved using relatively simple neighborhood operations, deconvolution and thus sharpening of the image is a more complex task. Neither point nor neighborhood operations change an image in terms of its size or its basic structure. This is achieved through geometrical image operations such as the scaling , rotation or translation of an image, anisotropic filtering being necessary here and interpolation being a decisive criterion for the image quality. The geometric image operations are part of the global image operations that use the complete image as input data. Another representative of global image operations is the Fourier transformation , whereby the image is converted in the frequency space in which the use of linear filters means less effort.
The time required for the operations mentioned is very much dependent on the image resolution .
Examples
Image enhancement
Original image
Image with artificially added salt-and-pepper noise.
Applying a 3 × 3 Gaussian filter to the noisy image.
Applying a 3 × 3 median filter to the noisy image.
Object detection and tracking
Flying object detection through image processing
Detection of moving people through Shape Recognition image processing
See also
literature
- Wilhelm Burger, Mark J. Burge: Digital Image Processing: An Algorithmic Introduction Using Java . First edition. Springer, 2008, ISBN 978-1-84628-379-6 .
- Helge Moritz: Lexicon of image processing . Hüthig, Heidelberg 2003, ISBN 3-7785-2920-X .
- Bernd Jähne : Digital image processing . 6th revised and expanded edition. Springer, Berlin a. a. 2005, ISBN 3-540-24999-0 .
- Pedram Azad, Tilo Gockel, Rüdiger Dillmann: Computer Vision. The practice book . Elektor, Aachen 2007, ISBN 978-3-89576-165-2 .
- Gerhard A. Weissler (Ed.): Introduction to industrial image processing (= electronics & electrical engineering library . Volume 1. ). Franzis, Poing 2007, ISBN 978-3-7723-4028-4 .
- Michael Sackewitz (Ed.): Handbook for industrial image processing. Quality assurance in practice, 3rd completely revised and updated edition . 3. Edition. Fraunhofer Verlag Stuttgart, Stuttgart 2017, ISBN 978-3-8396-1226-2 .
- Johannes Steinmüller: From image processing to the spatial interpretation of images . 1st edition. Springer, Berlin / Heidelberg 2007, ISBN 978-3-540-79742-5 .
- Kristian Bredies, Dirk Lorenz: Mathematical Image Processing. Introduction to basics and modern theory . Vieweg + Teubner, Wiesbaden 2011, ISBN 978-3-8348-1037-3 .
- Rafael C. Gonzalez & Richard E. Woods (Eds.): Digital Image Processing . Pearson Education, New Jersey 2008, ISBN 978-0-13-168728-8 .
- Michael Sackewitz (Ed.): Guide to industrial image processing (Volume 13). Fraunhofer-Verlag Stuttgart, 2013, ISBN 978-3-8396-0447-2