Novelty and Change Detection

Ever since the first digital computer was constructed and equipped with sufficient hardware to process large amounts of data, digital image processing has become an increasingly used and desirable technique in many different fields. This is why it is important to understand the theory behind it and master its applications. While the human eye often is a final judge for an image that is to be interpreted, an understanding of how visual perception works is crucial from an image processing perspective. How the mammalian visual system perceive information is basically divided in three major parts, namely object perception, pattern recognition and object recognition. A more recent field within image processing is novelty and change detection which concerns applications in a variety of areas primarily to detect changes in dynamic scenes such as satellite surveillance, object tracking, and also in different geological and medical contexts such as measures of land erosion and cell distribution. It is not always trivial to define when or if a change and a novelty coincide. In digital images, a change is considered to be something detectable assuming the changed area exceeds the size of one pixel, while for biological systems this no longer holds. Humans are subject to the change blindness phenomena which makes us overlook concurrent changes. A change is often thought of an elapse in time; hence motion is a requirement for changes and vice versa. There are two main approaches to change detection: pixel-based and region-based. Image differencing is a pixel based change detection method and by far the most common technique found in the literature. It is only applicable when the foreground (object) is moving against some static background, this to ensure sufficient spatial matching and allow background cancellation. In all the other cases, a vast collection of more or less successful techniques are available. It is often desirable to combine different change detection methods to achieve optimal effectiveness for the intended purpose. Morphological filters are sometimes used for video surveillance and real time applications while it offers a trade off between the fast pixel-based methods and the robust region-based methods. A general approach to change detection has proved to be optic flow analysis which can be thought of as a vector-field approximation of the retinal projection. By using this technique moving object can be extracted from many types of backgrounds. Novelty and change detection has proved to be an extensive field of research offering several interesting applications. However, while it is relatively new, it leaves no doubt that it in the future will reveal still hidden applications and bring forth new discoveries. Novelty and Change Detection 5 Preface ........................................................................................................................................................................................3 Acknowledgements..................................................................................................................................................................3 Abstract......................................................................................................................................................................................4.....................................................................................................................................................................................4