Image processing
Image processing is the a key technology to enable computer vision with a broad range of applications.
susan.savage (at) acreo.se , robert.forchheimer (at) isy.liu.se
Image processing is a form of signal processing where the input is an image, and the output can be either an image or a set of parameters related to the image. The term “image processing” usually refers to digital image processing, although analogue image processing can also be carried out. Digital image processing allows the use of complex algorithms for feature detection/extraction and algorithms involving 2D and 3D geometry, (also known as computer vision).
Edge and corner dectection
A simple form of image processing is Histogram Equalisation, which is a method of contrast adjustment using the image’s histogram. Through this adjustment, intensities can be better distributed, which can lead to improved visibility of detail in the image.
Feature detection involves methods where computing algorithms make a decision at every image point as to whether there is an image feature at that point or not. A large number of feature detectors have been developed. Examples of these are edge detection, and corner detection or more generally interest point detection.
Edges play an important roll in many applications of image processing, in particular for machine vision systems that analyze scenes of man-made objects under controlled illumination conditions. Edge detection refers to algorithms which aim at identifying points in a digital image at which the image brightness changes sharply or has discontinuities. If edge detection is successful, interpretation of the image is substantially simplified. However, edge extraction from more complicated images is often hampered by fragmentation, where edge segments can be missing or false edges can exist.
Corner detection, or more generally interest point detection, is frequently used in image matching and object recognition. A simple approach to corner detection in images is the use of correlation which indicates the strength of a relationship between local patterns, but this can be computationally expensive and is not optimal. Other techniques for matching and recognition are holistic techniques such as principle component analysis.
Local descriptors have proven to be very successful in applications such as object recognition and image retrieval. When applied to a location in an image, a local descriptor gives a set of values that describe the properties of a small area surrounding that location, for example mean intensity, variance, and local orientation.
A number of methods have been proposed for computing or estimating orientation from image data. These include filter-based methods, the use of a local polynomial approximation, and the use of various tensors such as structure, energy and boundary tensors.
One algorithm which is used in computer vision to detect and describe local features in images is scale-invariant feature transform (SIFT). SIFT features are based on the appearance of the object at particular interest points, and are fairly invariant to image scale and rotation. They are also robust to changes in illumination, noise, minor changes in viewpoint, and partial occlusion. Object identification occurs by matching against a (large) database of local features.
Face recognition
Over the past ten years, face recognition has become a popular research area in computer vision using visible wavelengths, and is one of the hottest applications of image analysis.
Face recognition can be carried out by a computer from a digital image or a video frame for automatic identification or verification of a person. One way of doing this is by comparing selected facial features from the image with corresponding features stored in a facial database. These features can be, for example,
- distance between the eyes
- width of the nose
- shape of the cheekbones
- length of the jawline.
A relatively new technique is 3D face recognition. 3D sensors are used to record information about the shape of a face. This then enables identification of distinctive features on a face, such as the shape of a nose or chin. One advantage of the technique is that it is not affected by changes in lighting, and can identify a face from a range of angles including a profile view.
The field of computer vision can still be regarded as immature and diverse. Many methods and applications are still at the level of basic research. However, computer vision is being included in more and more commercial products as part of the system, in areas such as security, medical imaging, and industrial process control, each of which has its own image processing requirements.