Removal of Artifacts To Improve Image Change Detection using DFrFT

Abstract

Image change detection is increasingly becoming one of the major areas of research in image processing. Detecting regions of change in multiple images of the same scene taken at different times is of widespread interest due to a large number of applications in agricultural surveys, urban studies, environmental monitoring, video surveillance, remote sensing, medical diagnosis and treatment, civil infrastructure, forest monitoring, underwater sensing and driver assistance systems. The changes may be due to object movement, insertion, deletion, removal or deformation, and the changes are usually affect the spectral signatures at same pixel locations of two sets of images of the same scene. A key issue in change detection is that it should include only significant changes not the insignificant due to artifacts like, illumination variation, partial translation, large daylight changes and shadowing effect etc. The removal of these artifacts helps in implementing this image change detection system in various applications, like remote sensing, video surveillance and civil infrastructure etc, more accurately. Although there are a number of methods, their applicability is restrained by limitation of the information they are evaluated upon, the type of image acquisition available and need of information to be retrieved after change detection etc. The present dissertation undertakes a study of image change detection using FrFT along with intensity normalization and thresholding. FrFT has been used as it provides extra degree of freedom to detect accurate changed regions. The use of intensity normalization and thresholding ensure that change is based on appearance or disappearance of objects only, with removal of above mentioned artifacts. Intensity normalization helps in making mean of mutitemporal images equal. Thresholding has been applied to classify pixels as changed or unchanged. In the end, gradient co-relation has been used for classifying the changes obtained from difference image depending upon the value of correlation coefficient. Change detection results have been analyzed, using precision and recall parameters values, by using three methods namely, DCT, DFrFT and presented method. Results have shown that DCT method is poorest among DFrFT and presented method. By calculation, very large values of recall value have been obtained for all image sets using presented method, it shows that desired objects are detected. The overall improvement in recall value is 0-34% to DFrFT method. However, precision value for presented method is 6- 82% more as compared to DFrFT method which means very small numbers of false regions have been detected.