Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE credits
When SKL (Swedish National Laboratory of Forensic Science) gets images to analyse, the task often is to identify a person or an object. In that work it’s important to find features. Unfortunately images from camera surveillances often contain a lot of noise, so details can be hard to discern.
A simple method to improve an image from a sequence is to compute the mean of some of the images. However, the part of interest in the image has to be in the same position in all images. Neither the camera nor the object can make a move between the images, and then all the movements are seen in the image instead.
In this report, one way to register some images so they look like each other in a special area is described. The image, which is showing most of the interesting object, is chosen to be a reference image. Then the transformation which describes the global movement between the images can be found.
The transformation maps one image to each other with interpolating. When the object is equals in all interesting images they can get together and create a better image. All the algorithms have been written in MATLAB and a Graphical User Interface was also developed. To use this program when an object has done local deforms the image can be divided and local transformations can then bee computed. In the end of the report some results are present.
2006. , 46 p.